Smart Eye Pro 5.0 - Center for Brain, Biology and Behavior

Transcription

Smart Eye Pro 5.0 - Center for Brain, Biology and Behavior
Smart Eye Pro 5.0
User manual
SMART EYE PRO
User manual
REVISION 249
08-07-02
 Smart Eye AB
Första Långgatan 28B • 413 27 Gothenburg • Sweden
Phone +46 31 60 61 60 • Fax +46 31 701 05 15
Support mail address:
[email protected]
Table of Contents
SMART EYE PRO.................................................................................................................................5
Licensing..............................................................................................................................................................5
Corneal reflection and Estimated Gaze...............................................................................................................5
Pupil detection or Iris Detection.........................................................................................................................6
Getting started.....................................................................................................................................................7
System Overview.................................................................................................................................................8
Hardware...................................................................................................................................................................................................8
Smart Eye Pro Software............................................................................................................................................................................8
SYSTEM SET-UP (HARDWARE)..........................................................................................................9
Camera Placement...............................................................................................................................................9
Camera placement for screen measurement..........................................................................................................................................10
IR Illumination....................................................................................................................................................10
Positioning of the IR Flashes..................................................................................................................................................................11
Power saving mode ................................................................................................................................................................................11
Aperture and Focus Settings.............................................................................................................................11
Camera orientation..................................................................................................................................................................................11
Adjusting the camera brightness (aperture)............................................................................................................................................11
Adjusting the camera focus.....................................................................................................................................................................12
Sound feedback......................................................................................................................................................................................12
SYSTEM SET-UP (SOFTWARE).........................................................................................................13
System Configuration .......................................................................................................................................14
The Camera Configuration Tab ..............................................................................................................................................................14
The Flash Positions Tab ........................................................................................................................................................................15
Camera Calibration............................................................................................................................................16
Verify Camera Calibration.................................................................................................................................17
Defining a Coordinate System...........................................................................................................................19
Defining Coordinate System Automatically.............................................................................................................................................20
Defining Coordinate System Manually....................................................................................................................................................21
Use Camera 1 as Coordinate System Origin .........................................................................................................................................22
Defining reference rotations for heading, pitch and roll...................................................................................22
Defining the reference rotation................................................................................................................................................................23
PERSONAL PROFILE.........................................................................................................................25
Components of a Personal Profile.....................................................................................................................25
Creating a Personal Profile................................................................................................................................26
Step 1: Taking snapshots..................................................................................................................................27
The Snapshot Dialog...............................................................................................................................................................................27
Automatic Feature Detection...................................................................................................................................................................28
Modes......................................................................................................................................................................................................28
Eye Center Snapshots............................................................................................................................................................................28
Step 2: Selecting Poses.....................................................................................................................................29
Removing poses.....................................................................................................................................................................................29
Marking eye center poses.......................................................................................................................................................................29
Step 3: Mark facial features..............................................................................................................................30
Correcting or modifying automatic feature detection..............................................................................................................................31
The Auto Correct button..........................................................................................................................................................................31
Marking the eye centers..........................................................................................................................................................................31
Placing the markers................................................................................................................................................................................31
Sub pixel marking....................................................................................................................................................................................31
Marking Tracking Pictures.......................................................................................................................................................................32
Marking Eye Center Calibration Pictures................................................................................................................................................33
Extra Feature Points...............................................................................................................................................................................33
Eye Lid Ranges.......................................................................................................................................................................................33
Iris and Eye Radius.................................................................................................................................................................................34
Eye to use...............................................................................................................................................................................................35
Appending Poses to a Personal Profile.............................................................................................................35
Removing Poses.................................................................................................................................................35
Requirements on a Profile.................................................................................................................................35
Using / Changing an Existing Profile.................................................................................................................35
Analyzing a Personal Profile..............................................................................................................................37
Head model.............................................................................................................................................................................................37
Eye Centers.............................................................................................................................................................................................38
Errors in the camera calibration..............................................................................................................................................................38
Profile Statistics.......................................................................................................................................................................................39
The Pose Rotations Box.........................................................................................................................................................................39
The Head model Information Box...........................................................................................................................................................39
Placing markers automatically..........................................................................................................................40
Make the cursor jump to the current feature....................................................................................................40
Gaze Calibration.................................................................................................................................................40
RUNNING THE SYSTEM...................................................................................................................42
Playback Control ...............................................................................................................................................42
The Action Menu................................................................................................................................................43
Start/Stop Tracking.................................................................................................................................................................................43
Creating a Recording..............................................................................................................................................................................43
Logging....................................................................................................................................................................................................43
The Views...........................................................................................................................................................44
The Image Source View..........................................................................................................................................................................44
The Values View.....................................................................................................................................................................................44
The Graph View.......................................................................................................................................................................................44
The 2D World View.................................................................................................................................................................................45
The 3D World View ................................................................................................................................................................................45
The Zone View .......................................................................................................................................................................................46
The Screen View ....................................................................................................................................................................................46
The System Setup Menu....................................................................................................................................46
Selecting Image Source..........................................................................................................................................................................46
Calibrate Cameras and Verify Camera Calibration.................................................................................................................................46
System Configuration..............................................................................................................................................................................46
World Model............................................................................................................................................................................................46
The Profile Menu................................................................................................................................................47
The Options Menu and the Settings Dialog.......................................................................................................47
The Tracking Tab....................................................................................................................................................................................47
Chessboard.............................................................................................................................................................................................48
IR Flash Time Out...................................................................................................................................................................................48
Communication.......................................................................................................................................................................................48
Can Output Settings................................................................................................................................................................................50
................................................................................................................................................................................................................50
CANCaseXL Configuration Utility............................................................................................................................................................51
Text Log Settings....................................................................................................................................................................................51
Time Synchronization..............................................................................................................................................................................52
World Coordinate System.......................................................................................................................................................................52
Gaze Filter ..............................................................................................................................................................................................52
The Toolbar........................................................................................................................................................52
Keyboard Shortcuts...........................................................................................................................................54
Main Application......................................................................................................................................................................................54
2D world model view...............................................................................................................................................................................54
3D world model view...............................................................................................................................................................................54
Zone view...............................................................................................................................................................................................55
Graph view..............................................................................................................................................................................................55
Profile dialog............................................................................................................................................................................................55
Playback Control ....................................................................................................................................................................................56
USING A WORLD MODEL .................................................................................................................57
World Model Syntax...........................................................................................................................................57
LocalCS ..................................................................................................................................................................................................57
Plane ......................................................................................................................................................................................................57
Screen ....................................................................................................................................................................................................58
Calibration Point......................................................................................................................................................................................59
Sphere.....................................................................................................................................................................................................59
Box..........................................................................................................................................................................................................59
Viewing the World...................................................................................................................................................................................60
Reload the World Model..........................................................................................................................................................................60
..............................................................................................................................................................................................................60
THE REMOTE CONTROL INTERFACE...............................................................................................61
Documentation........................................................................................................................................................................................61
Methods...................................................................................................................................................................................................61
Examples.................................................................................................................................................................................................61
FAQ / TROUBLESHOOTING...............................................................................................................62
General...............................................................................................................................................................62
Frame Loss.........................................................................................................................................................62
Logging / Communication..................................................................................................................................62
Camera Calibration............................................................................................................................................62
World Model / WCS ............................................................................................................................................63
Profile Generation..............................................................................................................................................63
Cameras.............................................................................................................................................................63
IR-flashes...........................................................................................................................................................63
SYSTEM SPECIFICATION.................................................................................................................64
Hardware............................................................................................................................................................64
Frame grabber.........................................................................................................................................................................................64
Cameras and Frame Rate.......................................................................................................................................................................64
IR Safety..................................................................................................................................................................................................64
External Processor..................................................................................................................................................................................64
Chessboard.............................................................................................................................................................................................64
Data Output Specification.................................................................................................................................65
Pupilometry........................................................................................................................................................70
Pupil filter.................................................................................................................................................................................................70
Quality ...............................................................................................................................................................70
Head Position Quality..............................................................................................................................................................................70
Gaze Direction Quality............................................................................................................................................................................70
Diagnosis............................................................................................................................................................72
Time Synchronization Specification (Optional module)....................................................................................72
Writing the dll..........................................................................................................................................................................................73
Logging Specifications......................................................................................................................................73
Communication Specifications..........................................................................................................................73
TCP/IP socket.........................................................................................................................................................................................73
UDP/IP socket.........................................................................................................................................................................................73
Packet Specification (TCP/IP, UDP/IP)...................................................................................................................................................74
Smart Eye Serial Format (RS232)..........................................................................................................................................................78
CAN Output Specification..................................................................................................................................79
Contact Information...........................................................................................................................................81
Smart Eye Pro
The Smart Eye Pro system is a head and gaze tracking system well suited for the demanding
environment of a vehicle cockpit and has the flexibility to cope with most research projects. The system
measures the subject’s head pose and gaze direction in full 3D. You can also receive eye lid opening
values and pupil dilation measurements depending on your needs.
The system can be used with up to six cameras with different lenses, allowing for a very large field of
view.
Licensing
Normally the license agreement allows you to install the software on one computer. Unless the system
comes pre-installed on your computer you need to create a small file, identifying your computer, and
email this to [email protected]. In return you will get a license file valid for your computer. The
license file also handles the unlocking of optional modules such as pupilometry or the remote control
interface.
The license information generation utility is always installed with smart eye pro and you can find it in
the start menu under Smart Eye Pro 5.0. To make a license information file, simply start the program
SmartEyeKeyGen and press OK and you will be prompted to name and save the file to an appropriate
location.
Corneal reflection and Estimated Gaze
As of Smart Eye Pro 4.0 we use the reflections of the IR flashes on the cornea (also referred to as
“glints”) to find the center of the eyes, rather than our old model where we estimated the eye center
using the head model. This allows for a more accurate gaze direction, and one that is less sensitive to
errors in the head pose estimation. These errors can still be induced by large head movements or
distorted facial expressions, but will no longer significantly impair the gaze direction.
Since we now use corneal reflection, the system is very sensitive to incorrectly positioned flashes so
you need to make sure that the flashes are defined correctly in the flash dialog.
To find the glints, the system also needs a clear, high resolution view of the iris, so depending on your
set up you might need to change lenses or move the cameras. The third difference from the old gaze is
that to build a gaze vector, each eye needs to be visible in two or more cameras.
Since these restrictions mean that you cannot use corneal reflection with all camera setups and all head
rotations, we always provide the old gaze vector as well. This is now referred to as Estimated Gaze.
If you have a system set up where you have problems with the Corneal Reflection, you can turn the
feature off in the settings menu. In this case, the old gaze direction will be found in the GazeDirection
value. This allows for full backwards compatibility with any client software you have.
If corneal reflection is turned off, Estimated Gaze and Gaze will be the same value.
All intersections calculated by the system will be calculated using the GazeDirection value, meaning
that no intersections are available for the estimated gaze direction except if you turn corneal reflection
off.
Note
Corneal reflection gives a higher accuracy and a more robust gaze vector, but it needs a high
quality image, and at least two cameras needs to see the eye at the same time. It is essential
that the IR flashes are correctly measured and that their position is set in the program.
5
Pupil detection or Iris Detection
The system is capable of using either the position of the iris, or the position of the pupil to calculate
gaze direction. The key differences is that the pupil position is more accurate, but, much similar to
corneal reflection vs head model based tracking, it might not be feasible to find the pupil in all setups.
If possible, we recommend that you use both pupil detection and corneal reflection as that will provide
a much higher accuracy. Moreover the use of pupil detection rather than iris detection will make the
system much more robust to individuals with high melanin content in the iris, which in IR, unfortunately
makes the contrast between the iris and the sclera very low. The good news is of course, that the pupil
will have a very large contrast in these cases.
In order to run pupil detection, you need to ensure a well focused, sufficiently high contrast image,
much like in the corneal reflection case. Furthermore: The system assumes a dark pupil, rather than a
bright pupil so you need to mount the IR-flashes at least 10 cm away from the camera. The easiest way
to ensure this is to use IR-flash mounts that are at least 10 cm long.
Note
The pupil gives a much more accurate gaze direction than the iris, but it is a bit more
demanding when it comes to image quality, and it also needs a flash placements that
minimizes the risk for bright pupil.
6
Getting started
To start measuring gaze, head pose or eye closure you need to perform the following tasks, all of which
will be described in detail below.
1) Assemble the system. Place the cameras in such a way that they won't be moved during tracking,
and make sure all cables are connected.
2) Perform a calibration of the cameras. This is an automatic process where you move a chessboard in
front of the cameras to calculate their relative position.
3) If you want to relate the measurements to the real world you need to create a World Coordinate
System (WCS). This can either be done automatically by placing a chessboard where you want the
coordinate system to be or manually by marking known real world points in each camera image.
4) The next step is to create a personal profile for your test subject. A personal profile is a collection
of snapshots with marked facial feature points. The profile is used to detect and track the face and
eyes.
5) The next step is to calibrate the gaze so the the gaze vector is aligned to the optical axis instead of
the visual axis. This is done by gazing at a number of known points.
You can then start the tracking by pressing 'F2' or selecting “Start Tracking” in the “Action”
menu. If you want to verify that the system operates correctly you can turn on “Values” in the
“View” menu.
If you want to send the data to one or more client application use the “Communication” tab in the
settings property page. Possible communication choices are TCP and UDP.
7
System Overview
The system consists of both hardware and software. The hardware components are mainly industry
standard high volume components.
Hardware
The Smart Eye system standard hardware consists of:
•
Standard desktop PC
•
PCI mounted frame grabber with adapter to the external processor
•
Two or more cameras with stand and cables
•
Two or more IR lights with cables
•
1-2 external processors with required cables and adapter
•
Chessboard for camera calibration
Optionally, it is possible to equip the Smart Eye system with up to five cameras and more than two IR
lights.
Smart Eye Pro Software
The Smart Eye Pro application is used for tracking and measurements as well as controlling the system
configuration.
FIGURE 1. Smart Eye Pro with 4 installed cameras.
8
System Set-Up (Hardware)
Camera Placement
Physically the cameras can be positioned totally independent from each other. However, in order to
maximize the performance of the system the following issues should be taken into consideration:
•
It is advantageous if each camera is oriented in such way that the subject’s head is at the center of
the camera image, while the subject is positioned in a neutral position. This will maximize the
allowable movement in x, y and z positions, without moving out of view of the cameras.
•
It is very important that the positions of the cameras and their orientation in respect to each other
are kept constant once the camera calibration process has been performed. Every time the cameras
are moved, you have to redo the camera calibration process.
•
In a vehicle environment one should strive to position and orient the cameras in a way that
minimizes the risk of direct sunlight straight into the camera.
•
It is of importance to position the cameras in such a manner that the hands and arms of the driver do
not obscure the field of view of the cameras when the steering wheel is spun.
•
To get as high quality data as possible it is important that the images are as zoomed in as possible,
so you should try to use lenses with as large focal length as possible and place the cameras as close
as possible that your wanted head box allows. To be able to use corneal reflection you also need to
make sure that the eyes are visible in two cameras at all times. To get good pupil detection values,
you need to ensure dark pupil rather than bright.
FIGURE 2. A four camera system mounted in an instrumented car.
9
Camera placement for screen measurement
FIGURE 3. A two camera system, mounted for screen measurement
In a (2 camera) screen measurement situation, the best setup for high accuracy is to place the cameras
symmetrically about 20-30 cm apart and 5-10 cm below the screen.
To get the best accuracy, use 12 mm lenses and our special pupil flash mounts. Make sure that pupil
detection and corneal reflection is turned on. By using the middle of the cameras as origin for your
WCS it will also be very easy to build the world model. If you mount the cameras and screen on a
screen stand it easy to adjust screen and cameras to accommodate subjects of different height. Both
screen stands and screen mounts are available for purchase from Smart Eye.
IR Illumination
The system uses IR-diodes to illuminate the face of the subject and to minimize the effect of varying
lighting conditions.
The IR lights are by default placed together with the cameras (one per camera). This is not an absolute
requirement and for test purposes the IR lights can be repositioned freely.
If you want to use the pupil positions, rather than the iris position to calculate gaze, or if you are
interested in the pupil dilation measure, the IR flashes need to be positioned at least 10 centimeters
from the optical axis to ensure dark pupil rather than bright pupil. See the pupilometry section for more
details.
While positioning the IR lights one should strive to minimize the effect of shadows (as seen from the
cameras) created by each IR light and also strive to create as even facial illumination as possible. The
easiest way of doing this is to position the lights directly on the “outside” of the cameras as shown in
Figure 2.
To minimize the reflexes when wearing eyeglasses, it is often advantageous to position the IR lights
(preferably along with the cameras) a good deal below the dominating gaze direction
10
Note
Do not reposition the IR lights unless there is a good reason for it, since this will affect the
performance of the system. The personal profile is especially sensitive to changes in the IR
illumination. Make sure you update information about flash positions in the software. See
below.
Positioning of the IR Flashes
You can move the flashes from their original positions if you wish, the system needs this information
however, to prevent reflexes from corrupting the iris detection. After you reposition the flashes, go to
the Flash Positions tab in the “System Setup→System Configuration” menu and enter their new
positions. See the section “System Setup” for additional information on this topic.
Power saving mode
To prevent the flashes from burning out prematurely they can automatically be turned off. To adjust the
time, go to the IR Flash Time Out tab in the “Options→Settings” menu and enter the new time,
or to turn off this feature. When the flashes is turned off a system message will appear and the flashes
will be kept dark until you press “Resume” in the message dialog.
FIGURE 4. The Power
Saving Flash dialog.
Aperture and Focus Settings
Before starting the software set-up procedure, it is of importance to check the aperture and focus
settings on the cameras. Start the application and open the aperture and focus dialog (System Setup>Cameras->Aperture and Focus dialog.
Camera orientation
The first step is to orient the cameras such, that the subject's face is positioned in the center
of each image. This can be checked by opening the focus dialog. The face should fill the
center box in the focus dialog.
Adjusting the camera brightness (aperture)
For adjusting the brightness, the subject should look towards the camera to be adjusted, such that his
face is seen in the center box of the window. On the top of this box, the gray value histogram is shown.
The optimal gray value histogram includes the full spectrum from low to high gray values, see Figure 5
(left). Note that the background is not completely black and thus the histogram usually starts with gray
values around 30.
If the image is too bright, there are many saturated pixels (gray value 255), see Figure 5 (mid).
Compare with the optimal case, where there are only few saturated pixels, caused by reflexions on the
glasses, on the nose tip, or on glints on the iris. Figure 5 (right) shows an example where the image is
too dark. There are no pixels with high gray-values and thus the contrast is lower compared to the
optimal setting.
The optimal setting is also indicated by the red bar on the left side of the box. The optimal setting is
reached when the length of this bar is in a (local) optimum. Clicking on the Reset button, resets the
current maximum value.
11
FIGURE 5. Aperture settings: The image to the left has a near optimal setting, while the middle image
is too bright and the right image too dark. Note the peak at the histogram in the middle image,
indicating that there are too many saturated pixels.
Adjusting the camera focus
When adjusting the focus, the subject should look towards the camera to be adjusted, such that his face
is seen in the center box of the window. Alternatively, a sheet of paper with text can be placed in this
position. In the same way as with the aperture, the blue bar on the right side of the window indicates the
quality of the focus setting. Again, since the measurement depends on the image content, the optimal
setting might corresponds to a local maximum of the length of the bar and not necessarily to a global
maximum. Clicking on the Reset button, resets the current maximum value.
Sound feedback
Both, the aperture and the focus setting is supported by sound feedback, which is especially useful, if
the images are not seen when adjusting the cameras. The aperture is coded by a sound with low
frequency and the focus by a sound with high frequency, which can by activated or deactivated
separately by the corresponding check box. The volume of the sound codes the difference of the current
setting from the maximum measured since the last reset. Thus, the optimal setting is reached when the
sound is as quiet as possible. Since the optimum value might not reach the current maximum, the sound
might not disappear completely. The current maximum can be reset by clicking the Reset button.
Since the aperture has influence on the focus, it is recommended to first adjust the aperture (deactivate
sound feedback for focus) and after that adjust the focus.
12
System Set-Up (Software)
The system needs to be set up before the measuring can start. All setup is done in the ”System
Setup” menu. Usually you only have to do this once. However if you move your cameras, want your
values in a new coordinate system or want to select another world model you have to set up the system
again for the new environment. If you experience a sudden lapse of quality of your measurement your
cameras might have been moved. It is then advisable to control the camera calibration by use of the
“Verify Camera Calibration” dialog. If you have the cameras mounted in a place where the
risk of someone accidentally touching them is high, it's advisable to always do a camera calibration
before each test subject.
13
System Configuration
“System Setup→System Configuration” opens the dialog where you set up your system
configuration. Unless you change cameras, lenses or reposition your flashes you usually won't have to
change anything here.
This information is needed to compensate for lens distortion and for reflexes on the iris by the flashes.
Performance can be seriously impaired if this is not entered correctly.
The Camera Configuration Tab
Here you simply select the camera and lens that corresponds to your setup, and press enter when done.
Unless you have other values presented to you by Smart Eye the default values for your particular
camera and lens combination are what you should use.
FIGURE 6. The Camera Configuration Tab
14
The Flash Positions Tab
In the flash positions tab you enter the position for each flash. You can specify it in the coordinate
system for each camera or in the WCS. The coordinate system for each camera have the positive Y
direction upwards, the positive z direction straight out of the camera and the positive x direction to the
left. After adding a new flash or moving an existing one you can check their new position in the 3D
world view. The flashes are denoted by a yellow dot, and the cameras are denoted by a coordinate
system with Red Blue and Green axes. The colored bars symbolize the x-, y- and z-axes respectively on
the camera coordinate system.
FIGURE 7. The Flash Positions Tab
15
Camera Calibration
Note
Before calibrating the cameras, make sure that you have selected the correct lenses and the
correct chessboard size in Options->Settings. If you select the wrong size, all measured
values will be incorrect and tracking might not work at all!
Once the cameras have been fixed in the desired positions the camera calibration process should be
performed. During this simple procedure, the application automatically detects the current positions and
orientation of the cameras.
To calibrate the cameras perform the steps below:
1. Choose “System Setup→Cameras→Calibrate Cameras…”. This will bring
up the camera calibration dialog, showing the live images of the currently connected
cameras.
2. Hold the chessboard in front of the cameras to perform calibration. The system requires
the whole chessboard to be visible in at least two cameras simultaneously to accept a
certain chessboard position. Showing the red, green and blue bars on the chessboard
indicates an accepted chessboard position. This is indicated individually for the cameras.
The colored bars symbolize the x-, y- and z-axes respectively on the chessboard
coordinate system. If possible, try to make the chessboard visible in all cameras.
3. Move the chessboard around until all progress bars are filled and the OK button becomes
active.
If the chessboard is clearly visible, and you see the yellow dots notifying a detected chessboard. But the
application displays the text “Could not Estimate Pose”, a probable cause for this is that you have
incorrect intrinsic lens parameters, check the settings in System Setup->System Configuration.
FIGURE 8. The camera calibration dialog. The figure shows a three-camera Smart Eye Pro system.
Note
Each time the camera positions have been changed, a new camera calibration must be
performed. The chessboard has to be entirely visible in a camera for it to be detected.
16
Verify Camera Calibration
After calibrating the cameras you should verify that they are correctly calibrated. This is done by
selecting “System Setup→Cameras→Verify Camera Calibration…” or by pressing
'F3'. The operation is fairly quick and you should do this often to check if the cameras have been
moved. This is very important to keep system performance on a high level.
Hint
Make it a rule to always verify your camera calibration prior to any session with the system.
To verify the calibration you need a chessboard just like in the calibration step. In the dialog you will
note a small check box under each camera image. Here you select which camera or which cameras you
want as “ground truth”. The pose of the chessboard in the checked cameras are then projected into the
other images using the current camera calibration. The difference between the tracked board and the
projected one is then shown as a red error vector (scaled by a factor of ten). The average difference is
also displayed in each image. Ideally this should be below 0.5 (but since the value is in pixels it
depends on the type of lens and how far away the chessboard is) . If you get a large value and see long
error vectors you have perform the calibration step again.
Hint
To get as much information as possible from the verification dialog., check each camera
against the other(s). Do this by checking and unchecking the check boxes below each picture.
FIGURE 9. The verify camera calibration dialog. Here the error vectors are small, and the accuracy
in the second camera is at 0.299 pixels, which is an acceptable value.
17
FIGURE 10. The verify camera calibration dialog. Here the error vectors in the right camera are large. It is
apparent that the cameras have been moved. In this case the right camera was moved downwards.
18
Defining a Coordinate System
Definition of your coordinate system is by default done by the automatic chessboard procedure. You
can however select your preferred method in the “Coordinate System” tab in the options dialog.
You will not be able to start tracking until you have defined the Coordinate System you want your
values in. This is done by selecting System Setup->Coordinate System Define...' in the menu and
following the steps outlined below. If you prefer to use a camera as the origin and rotation you don't
have to do any further Coordinate System definition. The disadvantage of this is that camera one is
normally not aligned with your desired world coordinate system and there is no way to reconstruct a
world coordinate system exactly, if camera one has been moved. Therefore we recommend that you
either use the Manual definition or use the chessboard.
FIGURE 11. In the options dialog. you can select how you want to define your coordinate system.
19
Defining Coordinate System Automatically
If you have a two camera screen measurement system, you can check the check box at the top of the
property page. This means that your origin will be placed between camera 1 and 2 and that the rotation
of the WCS will be determined by a line between camera 1 and 2 and the y-rotation of the chessboard.
This means that you will get an easy origin to measure the screen from, and if you make sure you hold
the chessboard vertically it will be very easy to build your world model. If the cameras are fixed to the
screen, you can even move the screen around without having to make a new definition.
With the box unchecked you can choose any combination of the following origins and rotations.
Origin
Center of Chessboard
Position of one Camera
Midpoint between two
cameras.
Rotation
Rotation of chessboard
Rotation of one Camera
Line between two
Cameras (x-axis) and the
rotation of the
chessboard (y-axis).
Description
The middle of where you hold the chessboard in the define WCS
dialog.
The position of a certain camera, you have to specify which one.
The midpoint between two different cameras, if two cameras are
symmetrically placed, it will be very easy to measure an object that
are between them.
Description
The rotation of the chessboard when you hold it in the define WCS
dialog. This is most useful if you can fix the chessboard at the same
position each time.
The rotation of a certain camera, not recommended you might want to
realign the camera without remeasuring your world model
The coordinate system will be built using a line between two Cameras
(x-axis) and the rotation of the chessboard (y-axis). If you hold the
chessboard vertically or with the y-axis aligned exactly in the same
way as you want your WCS rotated and the cameras are symmetrically
placed and fixed in relation to your objects it will be easy to measure
them.
The orientation of the chessboard coordinate system is as follows:
Axis
X
Y
Z
Orientation
Parallel to chessboard with positive X to your left.
Parallel to chessboard with positive Y upwards.
Orthogonal to chessboard with positive Z towards the cameras.
Color
Red
Green
Blue
Table 1. Orientation of chessboard coordinate system.
Follow these steps to define coordinate system automatically:
1) Make sure your selected definition method is automatic.
2) Select “System Setup→Coordinate System Definition... or press the
define button in the toolbar.
3) Choose which cameras you want to use by checking the respective box.
4) Hold the chessboard steady in the position where you wish to define the origin of the
coordinate system. Be sure to hold the chessboard in the “landscape”-position and not the
“portrait”-position.
5) When the system has calculated the pose of the chessboard, the OK button will become active.
Press the OK button to define the coordinate system.
The main advantage of this method is that it's very quick, you simply hold a chessboard in front of the
camera and when you press 'OK' the origin of the chessboard will be the center of the chessboard. The
20
disadvantage is that it is normally difficult to reproduce an identical chessboard position if you need to
redo the coordinate system definition. When using chessboard rotation and origin for your WCS it is
recommended that the chessboard is placed on a fixture to ensure that the Coordinate System is placed
on the same position each time you need to redefine it. If your objects are placed close to the cameras
(such as a screen between them), it is easier to use the midpoint between the cameras and only use the
chessboard to define the y-axis of the coordinate system.
FIGURE 12. Automatic Coordinate System Detection. Picture shows a 2-camera system.
Note
If you want to use the rotation and origin of the chessboard method as your main method for
defining the world coordinate system, we recommend that you mount the chessboard on a
fixture to make it easy to place the chessboard in the same position every time you need to redefine the coordinate system. It is necessary to re-define the coordinate system each time the
cameras have been moved.
Defining Coordinate System Manually
Selecting “System Setup→Coordinate System Define…” when your selected method is
manual pens a dialog where you can add landmarks in the landmark list and enter their coordinates in
the real world (the coordinates are always in meters). If you want to inspect or change a defined
landmark simply double-click on the landmark of interest in the landmark list.
When landmarks are defined in the landmark list; select the desired landmark, position the cursor over
the corresponding point (if visible) in the camera image and left-click in the correct position to connect
that pixel to the landmark coordinates. Right click to delete the connection.
Note
When you add a new real world reference point the coordinates should always be in meters.
21
FIGURE 13. Define Coordinate System. Picture shows a 2-camera system. This picture shows that landmark
“p6” is badly marked since the back projected point (in cyan) is quite far away from the marked point (in
green).
You will need to mark at least four landmarks in each camera. These landmarks need not be the same in
different cameras. You can check if the coordinate system is correctly calculated by checking the
“Show Back projection” check box. The landmarks are then recalculated into the camera
images and shown in cyan. If everything is correct they should be in roughly the same place as the ones
you marked. To make marking easier you can check the “Show Magnifying Glass” - button that
will magnify an area around the hair-cross. Check “Noise Reduction” to make it easier to identify
far away landmarks in the picture.
Note
Use well defined points for the landmark, e.g. mark points on the wall behind the system. If
the points are hard to see in the picture, you can put a drawing pin (with a metallic surface) on
the landmarks to improve visibility.
Use Camera 1 as Coordinate System Origin
By selecting “Camera 1” in the Coordinate System tab in the options dialog, the world coordinate
system will be defined by the position of camera 1 and its orientation with the x- and y-axis in the CCD
image sensor plane, the x-axis pointing to the right being parallel to the horizontal rows of pixels, the yaxis pointing up and being parallel to the vertical pixel columns and finally the z-axis starting at the
center of the CCD image sensor and pointing towards the object.
Defining reference rotations for heading, pitch and roll
Head rotation and gaze direction may be put out from the system in the form heading, pitch and roll,
also know as Euler angles. One may think of the heading as the left/right-rotation of the head, the pitch
as the up/down-rotation of the head and the roll as the tilt rotation of the head. Analogue for the gaze
22
vector, but roll is not applicable as the gaze direction is a vector and not a rotation. (The Smart Eye
system can not measure the torsion of the eyes.)
Any rotation may be described using Euler angles, but it is crucial that the heading, pitch and roll
rotations are applied in correct order.
In order for the Euler angles to be useful one need to define a reference rotation that defines the rotation
for which all Euler angles are zero.
The rotation matrix of the head pose has three base vectors. The base vectors originates at the head
center, which is located at the midpoint between the ears of the subject. The x-vector of the head
rotation matrix is the vector pointing out of the left ear of the subject, the y-vector is the vector pointing
'straight' up from the head center and finally the z-vector is the vector pointing forward out of the nose
of the subject. You can see the vectors as green coordinate axes drawn on the subject's head when the
system is in tracking mode.
The reference rotation for the Euler angles are defined by specifying two vectors; reference forward
direction and reference up direction. When the z-vector of the head rotation is parallel to the reference
forward direction and the y-vector of the head rotation is parallel to the reference up direction, the
heading, pitch and roll angles are all zero.
Out of the two reference vectors one can also create the reference left ear direction.
As mentioned, any rotation may be described using Euler angles. To create a rotation out of Euler
angles, do the following.
0. Start at the reference orientation
1. Rotate -heading about reference up direction. We have defined that a positive heading will turn the
head to the right, therefore the minus.
2. Rotate -pitch about the new reference left ear direction, as rotated in step 1. We defined that a
positive pitch will turn the head upwards, therefore the minus.
3. Rotate +roll about the the new reference forward direction, as rotated by step 1 and 2. Positive roll
will give tilt to the right, therefore no minus.
Defining the reference rotation
The reference forward and up directions are defined under the tab Heading, Pitch and Roll. By default
the reference forward direction is set to z-axis and the reference up direction is set to the y-axis of the
world coordinate system. Therefore, if your world coordinate system is aligned quite well with the
typical zero rotation (neutral pose) of the subjects head, you will not need to alter the reference
directions.
FIGURE 14. The Heading, pitch and roll property page
23
Note
Make sure that the reference forward and up directions are never parallel. In that case
heading, pitch and roll can not be calculated and the system will set all to 0.0.
24
Personal Profile
A personal profile is made up of a group of pictures with a number of facial features marked in each
picture. These marked pictures are used to build a model of the head and eyes and the marked features
are also used as templates when performing feature tracking.
This text covers the general steps in creating a personal profile. Normally the profile creation process
needs to be customized slightly to suit the specific requirements of a certain study/system set-up.
In order to make the profile creating as simple and robust as possible there is also a separate guide for
creating the personal profile. The guide can be used as it is or can be customized to be used as a
guideline in a specific project.
Components of a Personal Profile
A profile is virtual model of the subject's head. The profile contains information about 3D location of
all facial features such as eyebrows, ears, eye centers and mouth. This part of the profile is called the
head model. It also contains small snapshots/templates of each facial feature taken from different
angles and with different cameras. The templates are then used in the tracking to locate the facial
features in the stream of pictures from the cameras. Based on the location of features in the pictures and
the 3D model of the head, the system calculates the head pose of the subject.
In order to create the profile a number of snapshots of the subject is taken and the facial features are
marked in all the snapshots from the different cameras. This process will be more and more automatic
over time.
An important component of the profile is the location of the eye centers. This location is highly
individual and needs to be determined for each subject. Determining the eye centers is done using and
marking separate snapshots, but is performed in parallel to creating the profile for pose tracking.
The profile thus consists of marked pictures that serve one of two purposes.
•
Eye center calibration: these pictures are used to calculate the left and right eye centers of the
subject. The marked features will not be used when tracking, but only for computing the head
model and the head pose in each snapshot. In addition to marking the facial features, the center
of the irises should also be marked for these snapshots. A correct eye center is crucial for the
accuracy of the system as it is used together with the tracked irises to calculate the gaze vector.
•
Feature tracking: normally a set of pictures (snapshots) evenly spread along the dominating
head rotation direction, e.g. head and gaze pointing straight forward with respect to the
cameras, 20 and 40 degrees respectively horizontally to the left, then 20 and 40 degrees
horizontally right. It is not recommended to have differences in rotations larger than 20
degrees. Head orientations of special interest can also be added, e.g. head and gaze pointing to
the rear view mirrors and/or to the radio/speedometer/climate control unit in a car. The feature
points marked in tracking snapshots will be used for tracking, therefore it is of great
importance how the feature looks, when making the decision about marking it or not. More
details on this will follow.
Note
•
•
A profile pose consists of the pictures for each camera taken at the same point in
time. On a three camera system, each pose will be a set of three corresponding
images. Each profile pose also corresponds to a specific head pose of the subject,
which explains the term pose.
The head model is a three dimensional model of the head of the subject. All marked
features in the profile will be part of the head model. By detecting the positions of
some of the features in the head model while tracking, the system may calculate the
pose of the head and thus the positions of all other features in the head model.
25
•
To calculate the head pose from a profile pose a couple of features needs to be
marked in the corresponding pictures. The system decides when a head pose can be
calculated for a specific profile pose. Poses not used are marked in the image list.
Creating a Personal Profile
To create a new profile, choose “Profile→Create New...”. Then a “Profile Wizard” dialog
appears where you follow three steps:
1) Take snapshots, this can be done in a couple of different ways.
2) Check the snapshots that were taken, remove snapshot and mark eye center snapshots
3) Check and mark the facial features in the selected poses.
26
Step 1: Taking snapshots
This step is for taking snapshots. It is important to use a clear and well prepared process for taking the
profile snapshots, otherwise the subject may get confused and nervous. It may prove convenient to use
physical points in the world to gaze at for the different snapshot directions.
The Snapshot Dialog
FIGURE 15.The Snapshot Dialog
The Camera button or 'S' adds a feature tracking pose to your selected poses
The Eye button or 'E' adds an Eye Center pose to your selected poses
The Add from Bookmarks button is only enabled when running a recording and the No Feature
Detection option is marked. It is not available when using live cameras. The button adds all the
bookmarks in current the recording to the profile.
There are three different modes of operation available:
●
Feature Detection On – Automatic Snapshots
●
Feature Detection On – Manual Snapshots
●
No Feature Detection
No Feature Detection is the default value.
If automatic feature detection is on the graph in the center of the dialog displays the head rotation of the
taken snapshots and the current rotation.
27
Automatic Feature Detection
Automatic feature detection is new for this version of Smart Eye Pro. Markers for facial feature points
can be positioned automatically when creating a new profile. At the moment it is only designed to work
for two cameras but in certain situations it might work with more than two cameras.
The automatic initialization uses elastic face models to find corresponding facial features in both
images. In the beginning, the user should look straight ahead, until the models in both images are fitted
correctly to the face. The corresponding 3D face model and the head pose are computed continuously.
The 2D face models are painted into the image with green dots, while the back projection of the 3D
model is painted with white dots. With the 2D face models tracking the face, the user can rotate his
head slowly to both sides, while snapshots are taken continuously if all 2D models agree in their marker
positions. If the tracking fails, the user should look straight ahead, until the models reinitialize.
There are several restrictions that have to be kept for the automatic profile initialization to be
successful.
Camera setup
The current version of the automatic initialization supports two cameras positioned with a distance
between them of about 20 to 50 cm. The cameras should be placed below the area of interest, for
example below the screen for a screen measurement set up.
Visibility
The face from eyebrow to chin as well as the ears have to be completely visible in both images for the
models to be able to converge. If the subject has long hair covering the ears, it should be removed.
Head movements
During automatic initialization, the subject should not move his head abruptly as this could cause the
2D face models to lose tracking. When varying the head pose, this should be done with rather slow and
smooth movements.
Facial expressions
The flexible face model can handle facial expressions to some extent, but during automatic profile
initialization, facial expressions should be avoided.
Modes
The mode Feature Detection On – Automatic Snapshots takes snapshots continuously while the
subject moves the head. New snapshots are only added if the distance to previously taken snapshots is
greater than a certain threshold. This mode doesn't take any Eye center snapshots so these should either
be taken manually by pressing the Eye center button or by marking a snapshot in the next step.
Feature Detection On – Manual Snapshots also uses the automatic feature detection but taking
snapshots is left to the user. Normal snapshots can only be taken when the dot indicating head rotation
is green, Eye center snapshots can be taken at any time.
Selecting No Feature Detection turns of the automatic feature detection. This means that all snapshots
taken needs to be marked manually in step 3.
Eye Center Snapshots
Take snapshots for the eye center calibration, by pressing the Eye button. This is of great importance,
please make sure that you understand the discussion below, otherwise contact Smart Eye. It is of
importance that the gaze direction/eye rotation relative to the head differs significantly for each
eye center calibration snapshot. This will make the calculation of the eye centers more accurate. The
best way of taking eye center calibration snapshots is to have a fix head pose for all the snapshots and
directing the gaze into the different cameras. This will give acceptable accuracy as long as the
difference in gaze direction relative to the head differs significantly between the snapshots.
A common mistake is to direct the head to one camera and also gaze into the same camera. Do not do
this, two snapshots like that will produce a very inaccurate eye center.
Hint
By using the camera or eye buttons, you can mark a snapshot as an Eye Center pose directly
from the Snapshot Dialog.
28
Step 2: Selecting Poses
Any snapshots you added in the previous step will be in the list of profile poses. If you have used the
automatic feature detection you should make sure that the feature points are marked at the correct
positions. In case of incorrect positions you can either remove the pose or proceed to the next step and
correct the positions.
FIGURE 16.The first Auto pose
Removing poses
If you for some reason do not want to keep a pose you can select the pose in the list and press remove
pose.
Marking eye center poses
If the pose is an eye center snapshot but the normal snapshot button was used you can use the check box
to mark it.
When all the relevant poses are selected and properly marked you go to the next step of the wizard by
pressing “Next”.
29
Step 3: Mark facial features
This dialog is used for marking features in the previously selected pictures. All features you can mark
are in the list labeled “Feature points” to the left in the dialog. You select the feature you want
to mark by clicking on it in the list or by pressing the “down” or “up” arrows.
You browse through the pictures by using the “left” or “right” arrow keys or by clicking t the
thumbnails below the current picture.
If you press “control” when using the arrow keys you will automatically jump to the same pose in
another camera. The buttons to the left and right of the pose name also takes you to the previous or next
pose
A marker is placed on a feature point in a picture by moving the mouse over it and clicking the left
mouse button. A right click removes the selected marker.
There’s also a window around the cursor that provides a zoomed contrast enhanced image. You can
turn this feature on or off by pressing 'ctrl-m' or by selecting “View->Magnifying Glass” in
the menu. The magnifying factor can be changed by using the ´+´ or ´-´keys.
FIGURE 17. The Place Markers Dialog,.
Note
Camera images are mirrored, so a point on the left side of the head such as “left
mouth corner” will be displayed to the left in the image. Think of it as “left is
always left”
30
The placing of markers in a picture is done a bit differently depending on if the picture is to be used for
tracking or for an eye center calibration. The tracking pictures are those you took of your face moving
or you gazing directly at an object of interest while the eye center pictures are of you gazing directly
into a camera.
Correcting or modifying automatic feature detection
When using the automatic feature detection the left and right ears are not marked. To get the best
tracking performance you should make sure to mark them manually. In some cases the automatic
feature detection get the eye corners a bit off, you should make sure that all eye corners are at the
correct position.
The Auto Correct button
The Auto Correct button uses the calculated head model and moves user marked features closer to the
calculated position to reduce errors. This should only be used when all the features in the Top Errors
list is at least orange. Pressing the button when having red members in the Top Errors list can produce
larger errors. The auto correct button only corrects the top ten errors in the list.
Marking the eye centers
It is a little misleading to call this procedure eye center marking as it actually is the iris center that
should be marked. The idea is however that the eye center will be right behind the iris center if the
subject is gazing straight into the camera. Indirectly it it thus the eye center that is marked.
It is common that users misinterpret the marking of eye centers, so please read following sentences
carefully.
The eye centers, i.e the iris centers, should be marked, only in the snapshot from the camera, which
the subject is gazing directly into and not in the corresponding snapshots from the other cameras. All
snapshots in the profile pose will be marked with the text eye center, but this is only to indicate that the
profile pose will be used only for determining eye center. All other features should be marked however.
If you have taken 2 snapshots (profile poses) for eye centers you should in total have marked 2 left eye
centers and 2 right eye centers. This is regardless of number of cameras. There will be 2 * number of
cameras pictures that are marked with the text eye center in the picture list. All these images should be
treated as eye center calibration pictures as they are all parts of eye center calibration profile poses.
Note
You can only mark eye centers in the pose you have previously marked as eye
centers (The text Eye Center will appear next to the pose and camera information
in the upper left corner of the image) . If you have omitted to do this, you can go
back one step in the process by pressing “<Previous” and check the check box in
the list of poses.
Placing the markers
A fast and relatively fool-proof method for marking the feature points is to click the same feature and
browse the images for all cameras and poses. You can use the left hand to navigate among the images
(left-right key) and the features (up-down key) and use the mouse to click. The left mouse button adds a
feature point, the right one removes a feature point. For example, start by clicking the left eye brow (if
that feature is prominent for the subject in question) while browsing through all images with the right
arrow. When you have reached the end of the image buffer, switch feature to the right eye brow and
traverse the image buffer in the reverse direction with the right arrow. You will probably find that
marking the features in this ”forward-backwards” fashion is a swift and efficient.
Please note that all features should not be marked in all images; some images are reserved for the the
eye centers (see below).
Sub pixel marking
Sub pixel marking mode can be entered by pressing and holding the left mouse button, or the shift key.
In this mode the magnifying glass freezes around the current mouse position, which allows you to mark
between pixels. Releasing the left mouse button inside the magnifying glass will mark the selected
feature in the position of the cursor.
31
The magnifying glass will appear every time you press the left mouse button, no matter if you have
turned it on or off, so if you don't need it for the time being, you can set the magnifying factor to 1 by
using '+' or '-' or by using the wheel on your mouse while pressing the left button.
FIGURE 18. Sub pixel marking simplifies the marking of eye centers significantly.
Marking Tracking Pictures
When marking the feature tracking points you need to keep in mind that some facial features are less
suitable than others. Due to the plasticity of the face, some features tend to move around more than
others, which can lead to errors in the estimation of the head- and consequently eye models. Depending
on your measuring situation you need to consider which features to omit.
The nostrils and the ears are always good features to mark. You also need at least one or two additional
feature pairs. You can add up to 20 extra tracking points. These can be placed on any part of the face,
or on anything that moves with the face such as glasses or markers, these, or some of the standard
features can be used.
When should I omit a certain feature in a particular snapshot?
•
If the feature is poorly visible or if the template box contains background, simply skip the
point.
•
Emphasis should be on contrast and visibility when marking feature points in order to ensure
good tracking.
•
Regarding the inner eye corner (for both eyes): these should not be marked if the point is
occluded or nearly occluded by the nose. No part of the silhouette of the nose should be in the
inner rectangle.
When should I omit a feature altogether for a certain test person?
•
The eyebrows should only be marked if they are highly visible with a contrast. If the
experiment requires high accuracy measurements in the vertical direction, the eye brows
should not be marked.
•
In order to achieve better accuracy in vertical direction, it is also good practice to skip both
inner and outer eye corners.
•
If the experiment requires that the subject be talking, if possible the mouth corners should not
be marked.
General Remarks:
•
The ear is a good feature if visible in the poses and cameras used in the experiment.
•
Always try to mark a feature in the same physical 3D position for all cameras and poses. This
can sometimes be tricky in cases where the feature appears in an extreme angle.
32
Hint
The square around the pointer shows the size of the template used in the tracking,
while the size of the magnifying glass corresponds to the search region. If there is
parts of background in the square you should not mark that particular feature.
Marking Eye Center Calibration Pictures
It is important to understand that no tracking templates are taken from the snapshots that are marked
with eye centers. Only the position is of relevance in calculating the head pose. Consequently, here you
should mark all visible features, regardless if the are poorly visible or not or if they contain a part of
background. Some features, such as eye corners and eye centers are essential.
Note
•
You should mark all visible features when marking the eye center calibration
pictures.
•
Make sure you mark eye corners carefully, even if you have chosen to not mark
them in the tracking snapshots. The positions of the eye corners are important in
detecting eyelids and irises.
•
Mark ears if possible, even if you have chosen to not mark them in the tracking
snapshots. The positions of the ears are used for determining the center of the head.
When all feature points have been marked in all the pictures (except the ones that should be left empty)
you finish the procedure by pressing the OK button. Finally you will be prompted to save the profile.
You can do this later if you wish but saving the profile it as soon as it is completed is recommended.
Extra Feature Points
You can add up to 20 extra tracking points. these can be placed on any part of the face, or on anything
that moves with the face such as glasses or markers. Just make sure that you place each point on the
same position in each picture.
If you use additional feature points you can cut down on some of the insecure features that tend to move
around a lot. Features such as eye corners, mouth corners and eyebrows are prone to move as the user
speaks, moves his eyelids or even moves his eye, and this makes them unsuitable for use in feature
tracking.
Just make sure that you mark the necessary features in the eye center calibration snapshots. In the eye
center pictures you should always mark eye corners. But it can often be a good idea to omit them in the
feature tracking images if you have enough extra features.
Note
Even if you choose to omit eye corners or ears in your feature tracking snapshots,
you should mark them in your eye calibration snapshots as they are needed for
detecting eyelids and irises.
Eye Lid Ranges
In order to prevent false answers from the iris detection or the eyelid detection you should calibrate the
maximum opening of the eyelids. In this simple procedure you simply set the upper and lower limits of
the eyelid opening. The system will only look for an iris in that region so it's important that they aren't
too small.
Select ”Eye Parameters->Adjust Individual Eye Settings” and use the sliders to
pick the limits. You will notice two half ellipses around the eye for the upper and lower limit. You
should make sure they fit good around the eyes, with a little extra room for extreme poses or a very
open eye.
The eyelid boundary is global for a specific profile, i.e. the same boundary is used in all poses. You can
navigate around the poses with the left and right arrows as usual to make sure the limits work well for
33
all poses. Press OK when you are done. The ranges are part of the profile and will be saved and loaded
together with the rest of your profile.
FIGURE 19. The adjust eye parameters dialog.
Iris and Eye Radius
The second use for this dialog is to change the radius for the iris and the eye. This can for example be
of use if a user has glasses that make the iris smaller than expected. Be careful with this feature
however, a faulty eye or iris radius will both make the iris detection less reliable and make the eye
vector point in the wrong direction.
Hint
A head model has to be created before you can see the eyelid ellipses.
34
When all feature points have been marked in all the pictures (except the ones that should be left empty)
you finish the procedure by pressing the Done button. Finally you will be prompted to save the profile.
You can do this later if you wish but saving the profile it as soon as it is completed is recommended.
Eye to use
If one eye is occluded or the user has a severe squint, you can select which eye you want to use for all
calculations here. The choice here reflects all calculations such as gaze origin, gaze direction, eye lid
opening and pupil dilation. This choice is valid for this particular profile only.
Appending Poses to a Personal Profile
You can add images to a previously created profile. This can for example be useful if you want to
perform an eye center calibration or if you note that you need more poses during tracking. Press
“<Previous” until the step 1:recording is reached.
Removing Poses
To remove one or more poses press “<Previous” to go back to step 2: Selecting poses. In this step
you can remove poses.
Requirements on a Profile
In order for the tracking to work a number of requirements for the personal profile has to be met.
The system automatically controls your profile while you are placing markers and you will note the text
“Can't build” in the Pose Rotations box to the left of the dialog (more on that in the “Analyze Your
Personal Profile” section below.
Currently the profile has to meet the following requirements:
•
Each marked feature point has to have at least 2 occurrences.
•
Each pose needs at least 4 marked features (although empty poses are allowed).
•
The poses have to be fully connected. A connection between two poses exists if they have at
least 3 features in common.
•
The features "eye corners", "nostrils" and "mouth-corners" always has to be marked.
•
There has to be at least one non-empty pose without any marked eye-centers.
Note
Although the requirements above will satisfy the formal conditions of the system, they will
not ensure good tracking. In order to achieve good tracking the above procedure for
recording and creating a profile should be followed and result in, typically, 5 different, well
spread (but less than 20 degrees apart), profile images along the dominant head rotation, with
marked features. For an accurate gaze measurement it is strongly recommended that an “Eye
Center Calibration Profile” (according to the above procedures) is made for each camera.
Using / Changing an Existing Profile
If an existing profile is to be used, choose “Profile→Open…” and open a profile of choice.
Note
A profile file consists of two separate files with the endings .sep and .seb respectively. Both
files have to reside in the same folder on the file system. If a profile is moved, both files must
be moved to the same location.
35
Your last four opened or saved profiles can be found in the “Profile→Recent Profiles” sub
menu.
You can also edit a previously created or loaded profile by selecting “Profile→Edit
Current...”.
36
Analyzing a Personal Profile
Head model
As soon as there is enough information to build a head model the system starts calculating it. You'll
note that the text in the Pose Rotations box changes to “Building” and shortly instead of the text you'll
see a pie chart representing the estimated rotations of the current model.
Once your head model is built you'll have a number of different ways to analyze the profile and correct
any possible mistakes you've made during the marking. The most obvious method is to turn on the
presentation of the head model in “View->Head model”. This displays the head model as small
plus-shaped markers for each feature in each picture. By simply comparing these to the markers you
have placed and the position of the features in the face you get get a general idea of how correct the
model is.
If the head model differs a lot from the actual head position there are three possibilities:
1) You need to mark more features.
2) The ones you have already marked are incorrectly placed.
3) There is something wrong with your camera calibration.
To check if markers are incorrectly placed you can select “View->Error Vectors”-. The
difference between a marked feature and the same feature in the head model will then be displayed as a
vector in the current picture. Longer vectors of course represents a large error.
Even if you have poses that are marked by the automatic feature detection it is imperative that you
check them for any misplaced features.
FIGURE 20. A decent head model.
FIGURE 21. The left eyebrow is incorrectly
placed which makes the whole head model shift
to the right.
Note
If you have moved or removed a marker or added or removed poses the system takes a short
while to regenerate the head model. Meanwhile the head model will be displayed in a
different color.
37
Eye Centers
There are basically two types of errors one can make when marking the eye centers, the first one is that
the marker isn't placed exactly in the middle of the iris, a convenient way of making sure the marker is
in a good position is to place it on the middle of an IR reflection (assuming there is an IR-flash close to
the camera you are gazing into.)
The second type of error is when the difference between the angle of the facial direction and the gaze
direction isn't large enough. A common mistake is to turn the head into the same camera as the one you
are gazing into when taking the eye center calibration snapshots. This will provide very little
information about the depth of the eye center, especially if all snapshots are taken in the same way.
This can easily be seen if you turn Show Head model on. The green arcs denoting the allowed eye lid
positions will often look very weird if the eye center is not correctly calculated. In the pictures below
the left picture is correct (the head is turned in another direction than the one you are gazing at). And
the green arcs fit nicely around the eyes. The right picture however, has no depth information since the
head is not turned. And thus the system has calculated the eye center far closer to the irises than it really
is. This will make the arcs become bloated. If they had been very small instead, it means that the
calculated eye center is too far away.
FIGURE 22. A correctly marked eye center. FIGURE 23. This eye center is not correct!
Errors in the camera calibration
If you experience strange errors that can not be explained by a misplaced feature, it is possible that there is a
problem with the camera calibration. This is especially likely if all error vectors in a particular camera points in
the same direction. To prevent these pesky errors, make sure you always calibrate your cameras before making
a personal profile.
The system can detect possible errors in the camera calibration. If a feature is marked on a position that
differs significantly from the expected position according to the camera calibration, it will be added as a
dark red error in the error list (see below). If you click this error in the list you will see a red text in the
camera image, describing the error and it's distribution over the cameras. You will also see the total
error distribution over all cameras. If a certain camera is responsible for a large proportion of the errors
it is very likely that you need to calibrate the cameras again and make a new profile.
38
Profile Statistics
A helpful tool in analyzing and improving your personal profile and the head model is the profile
statistics section to the left of the profile dialog. The information below will appear only if a head
model has been built successfully. The dark red errors however, will appear even if you don't have a
head model yet.
The Pose Rotations Box
In the first box you'll find the rotation of the poses around the y-axis. Each pose is represented by a line
with the neutral (the most forward facing) pose pointing upwards. This is useful if you want to make
sure your poses are at a reasonably even spread. If two poses are more than 20 degrees apart the area
between them will be yellow. And if they're at 25 degrees or more apart in the area will be red. For
optimal performance in rotations around the y-axis you should aim for all green areas as in the picture
below.
The Head model Information Box
Here you'll find the errors in pixels between your generated head model and the features you marked.
From left to right the errors are: The entire head model, The current pose, the current picture and the
selected feature.
Below you'll find a list of the features that differ the most from your generated model. The numbers
between the brackets represents camera and pose. The poses are sorted by error size, and also color
coded, where a red feature is over 5 pixels, an orange feature between 3 and 5 and a green one lower
than 3. Ideally all features should be green.
By clicking on a feature point you'll automatically skip to that feature in the correct camera and pose.
Using the list you can quickly verify that a feature is correctly marked and mark it again if necessary.
FIGURE 24. The profile statistics box, the little black dot represents the pose you have are currently
marking.
39
Placing markers automatically
When your head model starts to build you might notice that when there is enough information, the
projected head model (the plus shaped markers) are already placed in the correct positions without the
need to mark them by hand. If you wish you might just add a point there by turning “View>Suggest Point” on and then press “Space” to add a marker in the suggested position. You'll
note that an automatically placed marker looks a bit different (there's a circle around the cross), this is
simply to avoid confusion since these markers are not used in the head model generation. Using this
feature you can speed up the process of placing markers, but there is also a risk that features are
misplaced if not very careful.
Make the cursor jump to the current feature
If you turn “View->Suggest Point” off, you can instead use the space bar to make the cursor
jump to the feature you have currently selected. This can speed up the placing of markers significantly,
once you have a working head model.
Gaze Calibration
You calibrate the gaze by selecting ”Profile→Gaze Point Calibration”. In the dialog you
can select which predefined object to look at. The purpose of the gaze calibration is to determine the
difference between the visual and the optical axis of the eye.
Before you can start calibrating you need to add a number of calibration points. The points should be
easy to distinguish and evenly spread over your area of interest. If you are not using calibration points
on a screen or a plane, you need to know the 3D position of each point in the defined coordinate
system.
Simply add any points as calibration objects in your world model (see “Using a World Model”). If
your area of interest is a screen, you only have to define how many calibration points you want. The
system will automatically create calibration points for you on each screen. To display them on another
screen than the one smart eye pro runs on, you need a client that listens for the
CalibrationGazeIntersection value.
Load your world with the points and then select ”Profile→Gaze Point Calibration”. A
dialog will appear where you maneuver between the objects by using the “up” and “down” keys and
train for a specific object by looking at it and tapping “space”. The system will train until you tap space
again or until the progress bar is filled (after 1 second). Pressing “clear” or the delete button will clear
the selected sample.
40
FIGURE 25. The gaze calibration dialog
A sample will only be added if the quality of it is deemed good enough, but make sure you don't look
away or blink during the training because that will probably ruin your calibration. The red dot shows
where your current uncalibrated gaze intersects a plane, orthogonal to the current world point and a
vector pointing towards the eye center. The blue dots show all saved samples and the green dots show
the samples when the calibration algorithm has been run on them. Ideally the green dots should be in
the middle of the target. Carefully check that the blue dots are close together without outliers. If outliers
do exist, clear the current point and add new samples again. It doesn't matter if the blue dots are offset
from the target as long as the noise is small.
One circle in the target corresponds to +- 2 degrees of accuracy.
When the eyes are calibrated you can save your calibration in your personal profile. We suggest that
you give it a new name so you can use your old uncalibrated profile for reference.
For each point you can check the accuracy and standard deviation of the calibration, the accuracy will
depend on various parameters such as distance from the cameras, distance and position of the object
you look at and focal length of the lenses and individual differences. If any one point has a value that
significantly differs from the others, you should remove that point and redo the calibration.
41
Running the System
Playback Control
The playback control appears only when you are using a Smart Eye recording as your image source.
You can use it for navigating in a recording, and by using the bookmark frame feature, you can mark a
frame that you want to use for profile creation.
The slider consists of three markers, one is the start marker which sets the position where playback
starts. This positions frame number can be read in the first of the three text boxes on the left side.
There's also a stop marker which indicates the positions where playback stops or loops, the frame
number for the stop marker is displayed in the rightmost text box. Then there's the marker for the
current position, this can be moved to change the current playback position. The current frame number
is shown in the text box in the middle on the left side of the tool bar.
Bookmarks are displayed as a red line on the slider. Bookmarks can be used to create profiles, see
Profile Creation.
A brief description of buttons from left to right (Keyboard short cuts):
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Back - Returns the recording to the first frame or the frame marked with the green start
marker.
Step back - Moves the recording one frame back(H).
Play backwards - Plays the recording backwards.
Pause - Pauses the playback. Press again to resume(I).
Play - Normal playback.
Step forward - Moves the recording one frame forward(K).
Repeat - Toggles if playback will loop when reaching the start/end of the recording or the
start/stop markers.
Jump to previous Bookmark - Jumps to previous bookmark if available(CTRL+H).
Toggle Bookmark - Set or remove a bookmark(CTRL+J).
Jump to next Bookmark - Jumps to next bookmark if available(CTRL+K).
Save Bookmark - Saves the Bookmarks and the starts & stop markers to the recording.
Speed - Up increases playback speed, down decreases/reverses playback speed. It is not
possible to get slow motion playback(U & J). Higher speeds than 1x or lower than -1x means
that frames are skipped.
Zoom in - Expands the region between the green start marker and the red stop marker to
occupy the total length of the slider.
Zoom out - Resets the range of the slider to fit the entire recording.
Note
Playing backwards or skipping frames when tracking may result in loss of
tracking and strange tracking behavior. Proper tracking output is only
guaranteed with forward playback at normal speed.
42
The Action Menu
The “Action” menu contains three choices. The first is “Start/Stop Tracking” which turns
the tracking on and off. You need a personal profile and a calibrated Image Source to turn the Tracking
on. The second choice is “Record...” which is used to capture camera image to a file. You can then
select this file as your image source and use it for tracking or creating a profile. The last choice
concerns the logging of tracking data to file.
Start/Stop Tracking
Selecting “Action→Start Tracking” turns the tracking mode on and “Action→Stop
Tracking” turns it off. When the tracking mode is off, the system only shows the live images from
the selected image source.
Creating a Recording
To create a recording, first make sure that the application is not in tracking mode, and then choose
“Action→Record...”. The system will prompt you for a filename, when you have selected the
file the recording dialog appears. You can select the size of your recording to fit a CD, DVD or make it
arbitrarily long.
Since this function writes to the hard disk at a very high data rate, it is recommended that the disk is
defragmented and has plenty of free space left in order to avoid recording drop-outs.
Note
A Smart Eye recording consists of two separate files with the endings .sma and .smb
respectively. Both files have to reside in the same folder on the file system. If a recording is
moved, both files must be moved to the same location.
Logging
Smart Eye Pro has a logging function, which logs output from the system to a file. To start the log, first
make sure that the system is in tracking mode. Then choose a destination file by selecting
“Options→Log→Select Log file...”. When ready to start logging, select
“Options→Log→Log”. To stop the logging; unselect “Options→Log→Log”. Instead of using
the menu options, you may use the two buttons on the tool bar with the same functionality. Also see
“Logging Specifications” for a description of the log file format.
43
The Views
Each view has it's own pop-up menu that is displayed with a right-click of the mouse. Each of the six
views will be described in detail below.
The Image Source View
In the Image Source view the camera or recorded images are displayed, along with gaze vectors,
tracking points and other information. If your cameras are not connected, calibrated or if your
coordinate system is not defined a text will appear over the Images.
Right-clicking on the Image Source View window will make a menu appear Under this menu you can
select which type of information you want displayed in the Smart Eye Pro application window.
•
View->Gaze – In this sub menu you can select whether you want to displayed the gaze vectors and
if you want an individual vector for each eye or a consensus vector.
•
View->Eyelids – If you are measuring eye closure you can see the tracked eyelids as blue curves
around the eye.
•
View->Eye Clips – Will display a picture of each eye and a circle around it where the system thinks
the iris is positioned.
•
View->Head Orientation – This displays the head coordinate system in each camera.
•
Zoom – Here you can select if and how you want the images zoomed.
The old gaze will always be drawn in pink in the image view, while the gaze vector calculated with
corneal reflection is drawn in red. This means that with corneal reflection off, you will only get a pink
gaze vector.
Double clicking on an image in the image view, pops the focus and aperture dialog.
The Values View
This view displays the tracked values, such as head coordinates, gaze vector and eye lid opening. All
values are in your defined coordinate system. Right clicking on this view opens a menu where you can
either change the font of the view or select which values to show. You can display any of the values in
the Data Output Specification below.
The Graph View
The graph view displays a graph for the selected category, by enlarging the width of the graph view you
can examine values for a larger time window. Right-clicking on the view makes a menu appear, where
you can choose the value you want the graph view to show. Change settings such as which color the
graph should have or set the scale. Selecting auto set or pressing space bar will set the scale between the
highest and the lowest value since the last auto set.
You can also increment or decrement the scale by using a scroll mouse or the '+'or '-' keys. Using the up
and down arrow keys you can offset the graph up or down.
If a frame is lost, the graph view will simply skip a step, so if your graphs look very jagged you might
want to check your settings.
FIGURE 26. A Graph View showing eye lid opening state for the last 6 seconds
If you select color coded graphs, and the value you are looking at has a corresponding quality value, the
color of the graph will depend on the quality of the value that is displayed, for example for a head
44
position graph the color will change with the head position quality value. A red graph corresponds to a
low quality, while a green graph means that the quality is high.
The 2D World View
The 2D world view displays your world model in 2D if you have built and loaded one. If no world is
loaded or if you don't have the world module you can still see the camera and IR-flash positions relative
to each other. A camera is displayed as a coordinate system with the z-axis point straight out from the
middle of the CCD, while a flash bank is displayed as a yellow dot.
Right button opens a menu where you can select what you want to display and how. You turn on and off
the drawing of labels as well as different objects. You can also choose between X-Y, X-Z or Y-Z
perspective which gives you the option to look at your world from different angles.
FIGURE 27. A The same world shown in a 2D world view for each perspective.
You can rotate your world using the left mouse button and translate it by holding down left-ctrl and
using the mouse button to drag the world to your preferred position.
When you start tracking your head, eye center and a gaze vector will appear, when the gaze vector
intersects a world object the object will be highlighted in red and a blue dot will indicate the intersected
point.
The 3D World View
The 3D world view as indicated by it's name shows the world in 3D, as in 2D views it shows only the
camera and flash positions unless you have a world loaded.
FIGURE 28. A The same world as in figure 16 shown in a 3D world view.
You rotate the world by pressing the left mouse button and moving the mouse in the direction you want
to rotate. Translation is done in the same way, only you also press left control before pressing the
mouse button.
45
When tracking is started your head, eye center and a gaze vector will appear, when the gaze vector
intersects a world object the object will be highlighted in red and a blue dot will indicate the intersected
point.
The Zone View
The zone view displays a zone plane, in the right menu you can choose between each available zone
plane. As in the other views a zone that is intersected by the gaze vector will be highlighted in red and
the point of intersection will be denoted by a blue circle.
FIGURE 29. The first zone plane in the world displayed in the zone view.
The Screen View
The screen view draws the intersected dot on the actual screen, no matter where the window is
positioned. It can be used to verify the gaze intersection if you run Smart Eye Pro on the same screen as
the one you are looking at.
The System Setup Menu
The ”System Setup” menu is used to set up the system so it can start operating. Here you select
your preferred image source, calibrate your cameras if applicable and define the coordinate system you
want the tracking data to appear in. The world model is also selected here. If you don't move your
system and want to track under the same conditions you should only have to performs these operations
once.
Selecting Image Source
The system can use either the cameras or a recording as image source. The default image source is the
cameras. To change image source, the application must be in the non-tracking mode (toggle with
“Action→Tracking Stop”). Then select either “System Setup→Image
Source→Recording…” or “System Setup→Image Source→Cameras”. Selecting File
brings up a file dialog for selection of recording to use as image source.
Calibrate Cameras and Verify Camera Calibration
The camera menu also contains the choices for camera calibration and the verification dialog. These are
all described in detail in the 'Camera Calibration' and 'Verify Cameras Calibration' sections
respectively.
System Configuration
Contains the controls to select the type of lense and camera model you are using, as well as the
positions of the IR flashes in the world.
World Model
In the ”System Setup→Open World Model..” menu you select the world model you want to
use. You can also reload the currently loaded model and open it for editing if you have associated a text
editor to the world model files.
46
The Profile Menu
The ”Profile” menu is used for creating, editing, loading and saving a personal profile. This profile
is unique for each person and if the cameras are moved or the lighting conditions changed you need to
create a new profile. See the section about “Personal Profile” for more details.
This menu also contains the ”Gaze Calibration” dialog.
The Options Menu and the Settings Dialog
The options menu only contains one choice which to open the settings dialog. The dialog contains
choices like communication, logging and whether you want the system to track only head movements,
or if you want gaze or eye closure detection. It also contains the dialog for selecting a dll to amend a
user defined time stamp to the Smart Eye data output (Time Synchronization Module only)..
FIGURE 30. The Settings dialog
The Tracking Tab
Under the “Tracking” tab you can select between corneal reflection or head model based gaze
origin, and between pupil and iris based gaze direction. You will receive the highest accuracy if you use
both corneal reflection and pupil detection.
47
Chessboard
Here you can change the size of the chessboard, this is important to do when you change lenses as a
bigger chessboard might not fit with for example a 12 mm lens. If you select the small chessboard, and
use the larger one to calibrate or vice versa, all measured values will be incorrect! A good practice, if
you plan on using the same chessboard each time is to place a bit of tape over the chessboard you are
not using, which will prevent you from using it by mistake. It will also help you in the unusual but
potentially fatal case when both sides of the chessboard is visible in different cameras, but at the same
time.
IR Flash Time Out
Here you can set or turn off the IR time out for the flashes. This is simply a power saving feature to
make the flashes last longer. The flashes wont be turned off if you are in tracking mode.
Communication
The system is able to communicate with external systems in two different formats each with it's own tab
in the settings dialog: The UDP Settings and the TCP Settings tabs work analogously with the only
difference being that you have to define a destination host for the UDP connection.
FIGURE 31. The UDP settings property page
You can add as many connections as you wish. To add a connection just go to the corresponding tab,
and press the “Add New” button. A small dialog will appear where you specify host and port, and the
desired format. You can select between 3 different formats (two are older and for compatibility reasons
only). We recommend the more flexible Custom format. You can disable a connection temporarily by
unchecking the check box for that row.
48
FIGURE 32. The UDP destination host dialog
If you select the custom format you will also be required to specify which type of information you want.
Simply check the check box by the name of the value you want to send. See the “Data Output
Specification” section for more details about which values are available.
FIGURE 33. The Custom Output Specification dialog.
The specific format for the communication protocol is specified in the “Communication Specifications”
section. Free example code for a client receiving packets is available upon request from Smart Eye.
Contact Smart Eye for more information.
49
Can Output Settings
In the CAN log page you can turn can output on and off and select the base id for your can messages.
See CAN Output Specification for details on the different packets.
A check in the use-column means that the packet will be sent if CAN Output is enabled.
FIGURE 34. The CAN Output page: CAN Output is enabled, with base id 1024 (0x400), the selected
packets are CAN_SYNCH, and CAN_GAZEORIGIN and CAN_GAZEDIRECTION.
50
CANCaseXL Configuration Utility
The first time you turn CAN Output on and start the application the Vector Hardware Configuration
utility will start. In order for Smart Eye Pro to start sending CAN data you have to add an application
named “SmartEyePro” using Edit->Add Application in the Vector Hardware Config. Assign the
Channel on the CAN bus you want to use. (For more information on this, see the CANCaseXL
documentation). If you want to change the channel for Smart Eye Pro later on, you can access the utility
from the Control Panel.
FIGURE 35. Adding SmartEyePro as an application with the CANCaseXL Vector Hardware
configuration utility.
Text Log Settings
The text log settings is similar to the UDP and TCP settings pages. You simply select your desired
format, and if you prefer the custom format, you will have to specify the data you want. Your log files
will be in the format specified here.
51
FIGURE 36. The Text Log Specification dialog.
Note
Changing the specification of your custom text log packet will not affect
your custom packets set for the UDP or TCP communication. All
custom packets can be contain different data if desired.
Time Synchronization
The optional time synchronization module enables Smart Eye Pro to read absolute time stamps from a
user defined source. The “User Time stamp” tab configures this time stamp source.
World Coordinate System
Here you can pick between a coordinate system defined by manually marking known 3D positions,
using a chessboard or by the position or rotation of the cameras.
Gaze Filter
Here you specify at which angle an eye movement will be classed as a saccade (saccades will be
removed from the filter) and how long the fixation filter should be in seconds. The filter will not affect
the 'raw' head and gaze tracking values in the log or communication packets. instead it is used for the
processed data, such as which zone you are looking at and where in that zone. You can also turn on the
filter in the image view and the world model views. In those cases it only affects the visualized eye
vector.
The Toolbar
The toolbar contains the most common functions, namely: Start/Stop Tracking, Use Cameras, Open
Recording, Create Recording, Create New Profile, Edit Profile, Open Profile, Adjust
Focus/Aperture, Calibrate Cameras, Verify Camera Calibration, Define Coordinate System,
Select Log file and Start/Stop Logging.
52
FIGURE 37. The toolbar
53
Keyboard Shortcuts
Main Application
You can use the following keyboard shortcuts to speed up your work in the main application window.
Shortcut
Alt + F4
F2
F3
F4
Menu Choice
Exit
Start/Stop Tracking
Calibrate Cameras
Verify Camera Calibration
F5
F6
Ctrl + R
Reload World Model
Start/Stop Logging
Create Recording
Alt + F7
Alt + C
Alt + R
Options -> Settings
Image Source -> Cameras
Image Source -> Recording
Shift + [0..7]
Image Source -> Recent
Ctrl + D
Define your Coordinate
System.
Ctrl + W
Ctrl + O
Load World model
Open Profile
Ctrl + S
Ctrl + N
Save Profile
Create New Profile
Ctrl + E
Edit Profile
Ctrl + G
Shift + I
Shift + V
Shift + G
Shift + W
Shift + D
Shift + Z
Gaze Calibration
New Image Source View
New values view
New graph view
New 2D world view
New 3D world view
New zone view
Description
Exits the application.
Starts or Stops the tracking.
Detects and saves the position of the
cameras.
Verifies that the camera calibration
was successful.
Reload the current world model.
Start or stop the file logging.
Streams the image source to a file on
your hard drive.
Opens the settings dialog
Makes cameras your image source
Makes the recording of your choice
your image source
Makes one of your eight recently used
recordings your image source.
Opens a dialog where you can set the
origin of your coordinate system. If
you have selected 'Camera 1 ' as your
preferred coordinate systems this
choice is not possible.
Load a world model.
Open a previously created profile.
Save a profile to disk.
Create a new profile using the profile
creation wizard.
Edit your current profile, add or
remove poses and markers.
Calibrate the gaze
Opens an image source view
Opens a values view
Opens a graph view
Opens a 2D world view
Opens a 3D world view
Opens a zone view
2D world model view
Shortcut
1
2
3
+
Ctrl
Description
Perspective X-Y
Perspective Y-Z
Perspective Z-X
Zoom in
Zoom out
Enables translation by click and move instead of rotation
3D world model view
These shortcuts are only used for the world model orientation views. Use the num pad keys to navigate
around the 3D view (the lower right of the four views). And the '+' and '-' keys to zoom all views in or
out.
54
Shortcut
Num Pad 8
Num Pad 2
Num Pad 4
Num Pad 6
Num Pad 7
Num Pad 3
Num Pad 9
Num Pad 1
+
Ctrl
Description
Rotate up
Rotate down
Rotate left
Rotate right
Rotate left/up
Rotate right/down
Rotate right/up
Rotate left/down
Zoom in
Zoom out
Enables translation by click and move instead of rotation
Zone view
Shortcut
1..N
Description
Toggle between zone planes
Graph view
Shortcut
Up
Down
+
-
Description
Move the scale upwards
Move the scale downwards
Scale out
Scale in
Profile dialog
These shortcuts only work in the edit profile dialog.
Shortcut
Left
Right
Ctrl + Left
Menu Choice
Ctrl + Right
Up
Down
Ctrl + M
View -> Magnifying Glass
+
Ctrl + E
View -> Error Vectors
Ctrl + H
View -> Head model
Ctrl + P
View -> Suggest Point
Hold Shift / Hold Left
Button
-
55
Description
Browse pictures left
Browse pictures right
Jump to the previous picture in the
same pose
Jump to the next picture in the same
pose
Move to the previous landmark
Move to the next landmark
Displays a magnifying glass around the
cursor to simplify the placing of
markers.
Increment the magnifying factor by 1
Decrement the magnifying factor by 1
Shows or Hides error vectors between
the head model and the manually
placed landmark.
Displays the head model as plus
shaped markers for each marked
feature.
Turns the ability to automatically place
a marker on the corresponding head
model position. on or off.
Enter Sub pixel Marking Mode
Space bar
Automatically place a marker on the
corresponding head model position.
Ctrl + A
Ctrl + R
Ctrl + X
Poses -> Add New
Poses -> Remove
Markers -> Remove
Ctrl + L
Eyelids -> Set Eyelid Ranges
If suggest point is off, you will
automatically jump to the position of
the selected feature.
Add new poses to the profile.
Remove poses from the profile.
Remove all markers from the current
pose.
Set the upper and lower limit of the
eyelids.
Playback Control
These shortcuts can only be used when viewing a recording
Shortcut
H
I
K
Ctrl + H
Ctrl + J
Ctrl + K
U
J
Description
Moves the recording one frame back
Pauses the playback. Press again to resume
Moves the recording one frame forward
Jumps to previous bookmark if available
Set or remove a bookmark
Jumps to next bookmark if available
Increase playback speed
Decrease playback speed
56
Using a World Model
In order to find out which real world objects the user's gaze intersects a world model has to be defined.
The easiest way to do this is to use one of the examples in the Sample Worlds folder in your Smart Eye
Pro installation directory.
To build the the world model you need to create a text file with the ending .sew. You may use any text
editing software to create and edit you world files. We recommend that you use the free software
“Notepad++”, which can be downloaded from http://sourceforge.net/projects/notepad-plus/
in the Sample Worlds folder you will find an xml document called userdefine.xml if you place this in
your notepad++ directory (either in application data or the actual program directory depending on how
you configured notepad++, notepad++ will parse the world model syntax and highlight the allowed
keywords which greatly simplifies editing.
A world model can be loaded by selecting ”System Setup→Open World Model...”.
Note
All distances in the world model and elsewhere are in meters except when something else is
specified.
World Model Syntax
When placing a 3D object (LocalCS, Plane, Screen, Box, Sphere or Calibration Point) it should be
defined in world coordinates, unless you place it inside the brackets of a LocalCS. In that case the
position of the object should be measured from the origin of the LocalCS and with respect to the
orientation of your local coordinate system. Coordinate Systems can be nested, and you should always
use the coordinate system that contains your current object.
2D objects (Rectangular or Circular Zones or 2D Calibration points) can only be placed inside a Plane
or a Screen. In this case they should be defined in object coordinates, which for a Plane will be in
meters from the lower left corner of the plane, with positive x to the right and positive y upwards, and
for a Screen will be in pixels from the upper left corner of the screen with positive x right and positive y
downwards.
World models can contain any of the following objects:
LocalCS
A right handed local coordinate system used to move and rotate a group of objects. You can place a
LocalCS in the world coordinate system, or inside another LocalCS.
A LocalCS is defined by an origin and x and y-axises in it's parent coordinate system, which can be
either the World Coordinate system or another Local Coordinate System.
All coordinate systems in Smart Eye Pro are right hand systems.
LocalCS : {
name = "localCS"
origin = 0.1, 0.1, 0
xAxis = 1, 0, 0
yAxis = 0, 0.8, 0.2
}
Plane
A 2D plane that can contain rectangular or circular zones. The plane can be rotated by defining the x
and y-axis' or by placing in in a local cs. If the gaze vector intersects a plane, both world coordinates
and object (2D) coordinates will be provided. The name of the plane, and any intersected zones will
also be provided in the intersection object.
The plane can also contain 2D calibration points for gaze calibration.
57
A plane is defined by a a point (in your parent coordinate system), a size and optionally two vectors that
defines the orientation of the plane. If you want to use a coordinate system for each plane instead and
rotate the planes that way, that is perfectly fine.
The zone planes in turn can contain either circular or rectangular zones. Those are defined by a point
(in the coordinate system of the plane) and a width and a height.
The plane below contains a rectangular zone, a circular zone and four calibration
points.
Plane : {
name = "MyPlane"
lowerMiddle = 0.0, -0.05, -0.3
xAxis = 1, 0, 0 // optional, if undefined it will be 1, 0 ,0
yAxis = 0, 0.5, 0.2 // optional, if undefined it will be 0, 1, 0
size = 1, 0.5
RectangularZone : {
name = "rectangular zone"
lowerLeft = 0.1, 0.15
width = 0.25
height = 0.1
}
CircularZone : {
name = "circular zone"
center = 0.75, 0.25
radius = 0.1
}
}
CalibrationPoint2D : {
name = "c1"
center = 0.25, 0.15
}
CalibrationPoint2D : {
name = "c2"
center = 0.75, 0.15
}
CalibrationPoint2D : {
name = "c3"
center = 0.25, 0.35
}
CalibrationPoint2D : {
name = "c4"
center = 0.75, 0.35
}
Screen
Similar to the plane, but object coordinates will be in pixels instead of meters. It can also contain
rectangular and circular zones, and automatically or manually defined 2D calibration points. The name
of the screen and the name of any intersected zones will be provided in the intersection object.
A screen is defined by a a point, a size (in your current coordinate system) and a resolution and
optionally two vectors that defines the orientation of the plane. You can also make the system place a
number of calibration points for you, which is recommended, you should use 4 points for best results..
You do this by defining the keyword calibrationPoints = 4.
The screen in turn can contain either circular or rectangular zones. Those are defined by a point (in the
coordinate system of the screen which means the are in pixels) and a width and a height.
Screen : {
name = "MyScreen"
lowerMiddle = 0.0, -0.015, 0.02
xAxis = 1, 0, 0 // optional
yAxis = 0, .25, 0.04 // optional
size = 0.34, 0.271
resolution = 1280, 1024
calibrationPoints = 4
CalibrationPoint2D : {
name = "c1"
center = 145, 120
}
58
CalibrationPoint2D : {
name = "c2"
center = 500, 200
}
CalibrationPoint2D : {
name = "c3"
center = 600, 500
}
RectangularZone : {
name = "rectangular zone"
lowerLeft = 20, 20
width = 250
height = 200
}
}
CircularZone : {
name = "circular zone"
center = 600, 400
radius = 100
}
Calibration Point
Can either be a 2D point, in which case it needs to be placed in a Screen or Plane, or a 3D point in
which case can be placed in the world coordinate system or inside a local coordinate system. A
calibration point is only used in the gaze calibration phase and can never be intersected by the gaze
vector.
A 2D point in a screen:
CalibrationPoint2D : {
name = "c1"
center = 145, 120
}
A 3D point
CalibrationPoint : {
name = "CP_1"
center = 0,0,0.4
}
Sphere
A sphere of a certain radius which can be intersected by a gaze, intersections provide the name of the
sphere and the position of the intersection in world coordinates.
Sphere :
name =
center
radius
}
{
"a sphere"
= 0,0.15,0.3
= 0.1
Box
A 3D box, intersections provide the name of the box and the position of the intersection in world
coordinates.
Box : {
name = "MyBox"
0,0,0
.1,0,0
.1,.1,0
0,.1,0
0,0,.1
.1,0,.1
.1,.1,.1
0,.1,.1
}
59
Note
When you define your world, all objects should be in the coordinate system you plan to use.
The coordinate system should be manually or automatically defined since it's very difficult to
relate the world to a movable object such as a camera.
Viewing the World
When you have a world model loaded you can display it in two ways. The first way is by the world
views where you see the objects in your world and the subject's head and gaze in 2 or 3 Dimensions.
Use the world views to control that your world is correctly built, since you can see both the cameras and
the origin of your coordinate system in the views you can quickly check that your positioning is correct.
In the 2D view you can quickly look at your world from the side top or front by pressing the keys 1-3 or
by using the right mouse button menu.
The other type of display is the zone view. It displays each zone plane and the defined zones in it. The
spot where the gaze intersects the zones will be marked by a blue dot.
Both those views are described above in the section 'Running the System'.
Reload the World Model
By pressing F5 or by selecting “System Setup→Reload World Model” You can quickly
reload the world model if you have done any changes. This is mainly useful when defining the world.
60
The Remote Control Interface
The optional module remote control interface allows you to control Smart Eye Pro from an external
application. You do this by issuing procedure calls in XML-RPC (for full documentation refer to http://
www.xmlrpc.com/), this means that your application can communicate with Smart Eye Pro no matter
what your programming language of choice is. To connect to the XML-RPC server, use port 8000 on
the machine that Smart Eye Pro is running on.
Documentation
To get the documentation on line, simply go to port 8000 on the host machine. i.e.
http://localhost:8000/ for a server that is on the computer you are currently using. If you are connecting
to a remote host, make sure that your firewall won't block the connection.
Methods
These are the methods currently available in the remote interface of Smart Eye Pro. All functions return
0 if successful, otherwise an error code.
Method
LoadProfile(fileName)
setLogFile(fileName)
startLog(start=True)
startTracking(start=True)
stopLog()
stopTracking()
system.listMethods()
system.methodHelp(method_n
ame)
Description
Loads profile using the
provided filename
Sets filename of the log
Starts or stops the log
Starts or stops tracking
Stops the log
Stops the tracking
Returns a list of the methods
supported by the server
Returns a string containing
documentation for the specified
method.
Examples
Error Code(s)
1 - no profile with the
provided name was found
1 – no filename of log
specified
1 – no profile
You can find sample c++ and python code for the remote control interface in the Smart Eye Pro API.
61
FAQ / Troubleshooting
General
Q: The images from one or more of the cameras are all black.
A: Make sure that both the cameras and the flash cables are plugged in. Check the aperture
settings for each camera as well.
Q: The first time I used the Smart Eye system the values looked fine, now I've noted a lapse
in quality of the performance of the system.
A: Most likely one of your cameras has been moved, this can be due to an instable fixture or
in some cases due to large differences in temperature that might affect the position of the
cameras. Verify this by running the camera verification dialog, and if the error vectors are
large, re-run the camera calibration.
Frame Loss
Q: How do I avoid frame loss (You notice that the frame rate in the values field isn't exactly
30 or 60 fps)
A: Shut down unnecessary image source views, cut down on the number of open graph views
and make sure that eye clips aren't drawn in image source views. Generally the more views and
more information you show on the screen the more risk you run of experiencing lost frames.
You can check the frame loss by looking at the frame rate field or by pressing and holding left
shift. If frames are lost you will see the % of lost frames in the lower right corner of the
application window.
Logging / Communication
Q: How do I convert the rodrigues vector in the log file or the Smart Eye Packet to a more
user friendly format?
A: In the program folder you'll note a folder named 'Examples', look for the method
rodriguesToMatrix in the Matrix.cpp file. Using this you can easily convert a rodrigues vector
to an ordinary rotation matrix.
Q: How do I use the eye vector to calculate which real world object a user is looking at?
A: You can either use our world model module, or write your own code. There is an example
of how to calculate the intersection between a vector and a plane in the 'Examples' folder.
Camera Calibration
Q: I can never seem to get good values when i verify my camera calibration
A: Make sure you have set the correct size of the chessboard and make sure you wiggle the
chessboard around. To make a reliable calibration the system needs to see the chessboard in
different angles in each camera.
62
World Model / WCS
Q: The back projected values in the manual WCS dialog are very far from the points I've
marked by hand. I've made sure all points are correctly defined.
A: Note that all values should be in meters, if you still get a strange result, try to move some
points slightly or add a new one.
Profile Generation
Q: The back projected points in the head model in the profile dialog are very far from the
points I've marked by hand. I've made sure all points are correctly defined.
A: Move some points slightly or add a new one.
Cameras
Q: What are the default dip switch and gain settings for the Sony XC-HR50 camera?
A: See the hardware manual.
IR-flashes
Q: How can i check if the IR-flashes works properly.
A: Turn down the aperture setting and use a CD-case as a mirror to inspect intensity and
evenness.
63
System Specification
Hardware
Frame grabber
The system uses one Matrox Meteor II / Multi Channel frame grabber to acquire images from up to
three cameras.
Cameras and Frame Rate
The cameras are normally identical and of the model SONY XC HR50. These cameras are normally run
at 30 or 60 fps with a resolution of 640*480 pixels but has the capability of frame rates up to 240 fps
with reduced resolution. (Older Smart Eye systems use SONY XC55 or XC55BB). The cameras are
monochrome and are sensitive in the visible and near IR-range. IR-filters are attached in front of the
lenses of the cameras. The IR-filters stop visible light and thus makes the system more tolerant to
varying light conditions.
IR Safety
IR lighting is used to illuminate the drivers face with IR light. The Smart Eye equipment complies to
the appropriate international standard; IEC 62471 "Photo-biological safety of lamps and lamp systems.
This standard contains the calculation rules for Maximum Permissible Exposure (MPE), that Smart Eye
complies with. Under normal operating conditions the Smart Eye system has an approximately 300-fold
safety margin relative to the MPE. Even under worst case evaluation the safety margin is comfortable.
External Processor
The processor is used to control the IR-lights and the camera synchronization. The processor PC-card
also serves as a splitter adapter/connector between the cameras and the frame grabber.
Chessboard
The chessboard is used for automatic detection of camera positions. The chessboard should consist of
5*7 full squares plus a fringe of truncated squares. The sides of the squares should be exactly 30 mm.
64
Data Output Specification
From version 3.0 quite a large number of data items can be produced by the Smart Eye Pro system. It is
also possible for the user to select which of the data items that should be included in the output from the
system, both for text logs and for TCP/UDP data packets.
Version 3.0 is backward compatible and you may want to use an older communication format with your
old client. However, for building new clients we strongly recommend to base the communication on the
new generic format. See the “Communication Specification” below for details.
The available data items in the generic format are specified in the header file SEOutputDataIds.h
included in the sample network client. Please refer to this file for data types and ids of the specified data
items.
The available data items are listed in the following table.
65
Data item
Unit
Frame number
-
Required
Module
Description
A sequential frame number, set by the image
capture subsystem. The frame number starts
counting from 0 at application start and
increments by +1 for each successfully captured
image. This means that the frame number of two
consecutive frames will always differ by 1 even
if the image capture subsystem should happen to
drop one or more frames in between.
If the frame numbers in the data output should
increment by more than +1 from one data record
to the next, then it is either an indication of
frame loss in the image processing subsystem,
probably due to CPU overload – or, in the case
of UDP communication, loss of network data
packets.
Timestamp
10-7 s
Estimated delay
10-7 s
Time synch
User time stamp
Time synch
Frame rate
User
defined
fps
Head position
m
Head position quality
Head rotation
rodrigues
Head nose direction
Head up direction
Head left ear direction
rad
N.B: The frame number should primarily be seen
as a simple and unique data record identifier and
can not be used to detect frame loss. The only
reliable way to determine frame loss is to check
the time stamp difference between two
consecutive data records.
A high-resolution time stamp, measured at the
midpoint of image exposure and starting to count
from 0 when the application is started. Since this
time stamp is based on the PC hardware highresolution clock, it should only be used to
measure time differences over relatively short
periods of time.
The estimated delay from the real-world event
(midpoint of image exposure) to the network
socket send() system call.
64 bits of user defined information, usually an
absolute-time time stamp and/or annotation data.
The current frame rate of the system in frames.
Should ideally be 30 or 60 frames per second
depending on system frame rate.
The head position in 3D given in the defined
world coordinate system.
0..1 (0 = no cameras are tracking, 0.5 = 1
camera, 1 = 2+ cameras); A value less than 1
means “not reliable”.
The head rotation in 3D given on Rodrigues1
format in the defined world coordinate system.
Unit vector defining the 'nose' direction.
Identical to the z-axis of the head rotation
matrix.
Unit vector defining the 'up' direction. Identical
to the y-axis of the head rotation matrix
Unit vector defining the 'left ear' direction.
Identical to the x-axis of the head rotation matrix
1
Rodrigues format is a convenient way of describing rotation: The length of the vector defines the rotation
angle about the axis defined by the vector it self. Be careful not to interpret the components of the rodrigues
vector as true rotations about the axes in the corresponding coordinate system. This can only be done if all
components are close to zero, otherwise not. You can easily convert this vector to a rotation matrix and
example code that does this is shipped with your system.
66
Head heading
rad
Head pitch
rad
Head roll
rad
The left/right-rotation of the head. Also known
as 'no'-rotation. Refer to discussion about Euler
angles above for more information.
The up/down-rotation of the head. Also known
as 'yes'-rotation. Refer to discussion about Euler
angles above for more information.
The tilt-rotation of the head. Also known as
'maybe'-rotation. Refer to discussion about Euler
angles above for more information.
Head rotation quality
Eye position
0..1 The same number as the head position
quality.
m
The virtual eye position (origin of gaze vector)
in 3D given in the defined world coordinate
system. Calculated as average position of the
physical eyes.
A unit vector originating in the virtual eye
position describing the direction of the gaze.
Calculated as the average of the two gaze vectors
originating at the physical eyes. If vergence is of
interest, the eyes should be handled separately
instead.
0..1 Depends on the quality of the iris detection.
Calculated as the average of the quality of the
both physical eyes. See also discussion below.
The position of the center of the left eye ball in
3D given in the defined world coordinate system.
The center of the iris of the left eye, where the
gaze vector originates.
A unit vector originating in the left gaze origin,
describing the direction of the gaze of the left
eye.
0..1 Depends on the quality of the iris detection
of left eye. (see discussion below)
The position of the center of the right eye ball in
3D given in the defined world coordinate system.
The center of the iris of the right eye, where the
gaze vector originates.
A unit vector originating in the left gaze origin,
describing the direction of the gaze of the right
eye.
0..1 Depends on the quality of the iris detection
of right eye. (see discussion below)
The left/right angle of the gaze vector. See
discussion on Euler angles above.
Gaze direction*
Gaze direction quality
Left eye position
m
Left gaze origin
m
Left gaze direction*
Left gaze direction
quality
Right eye position
m
Right gaze origin
m
Right gaze direction*
Right gaze direction
quality
Gaze heading*
rad
Gaze pitch*
rad
Eyelid opening
m
The up/down angle of the gaze vector. See
discussion on Euler angles above.
Eye closure
The average distance between the eyelids of both
eyes.
* There are two version of these data items, unfiltered and filtered. The filtered data items have the same name
but are preceded with Filtered.
†There are two versions of this item, bot regular and estimated. The estimated data items have the same name
but are preceded with Estimated. The estimated values are not as accurate as the regular ones but on the other
hand you get a value even if there are no Glints visible on the iris (for example when they are occluded by an
eyelid). When you have turned corneal reflection off, the estimated and regular values will be the same value.
67
Eyelid opening quality
Left eyelid opening
Left eyelid opening
quality
Right eyelid opening
Right eyelid opening
quality
Closest world
intersection
Eye closure
m
Eye closure
Eye closure
m
Eye closure
Eye closure
World
Filtered closest world
intersection
World
World intersections
World
Filtered world
intersections
Calibration
Intersection
Keyboard State
ASCII Keyboard State
Left Diagnosis
Right Diagnosis
Left Eye Glints Found
World
Normally in the range 0..1, some subjects may
have values larger than 1.0. The value depends
on how distinct the eyelids can be detected. The
range is individual and can not be compared
between subjects. Calculates as the average
quality of the both physical eyes.
The distance between the eyelids of the left eye.
Quality of left eyelid detection. See also
comment above on Eyelid opening quality
The distance between the eyelids of the right
eye.
Quality of left eyelid detection. See also
comment above on Eyelid opening quality
The closest intersection with any of the world
objects. The intersection information contains
name of object, intersection point in world
coordinates and intersection point in object
coordinates. E.g. position in zone-coordinates.
Each sub packet of this type starts with an
integer indicating the number of world
intersections contained in the sub packet. If there
are no gaze intersections with world objects for
the current frame this integer will be 0, otherwise
1.
Same as Closest world intersection, but the
intersections are calculated based on the filtered
gaze direction.
Analogue to closest, but contains the whole list
of intersected world objects. E.g., if the right
window of a car is intersected it may be
interesting to also find out if the right rear mirror
is intersected.
As with closest intersections, each sub packet of
this type starts with an integer indicating the
number of world intersections contained in the
sub packet. The difference is that in this case
there may be any number of intersections.
Same as World intersection, but the intersections
are calculated based on the filtered gaze
direction.
Only used to send the selected calibration object
to clients. Useful if you want to display a
calibration point on a screen when performing
the gaze calibration.
The keys that are currently pressed (A-Z, 0-9),
useful for marking events in log files. Please
note that due to limitations in a normal PCkeyboard, pressing more than two keys at a time
gives undefined output.
The ascii code of the first pressed key, as an
integer.
Gives a diagnosis via a value below zero if a
gaze vector cannot be calculated for the left eye.
Gives a diagnosis via a value below zero if a
gaze vector cannot be calculated for the right
eye.
How many glints that are found on the iris of the
left eye
68
Left Eye Glints Found
Pupil diameter*
How many glints that are found on the iris of the
right eye
m
Pupil diameter quality*
Left pupil diameter*
m
Left pupil diameter
quality*
Right pupil diameter*
Right pupil diameter
quality*
m
Pupilometry
The consensus diameter of the pupils of both
eyes. See section on pupilometry for more
details.
Pupilometry
The value is normalized and will be in the range
of [0.0, 1.0].
Pupilometry
The diameter of the left eye pupil.
Pupilometry
The value is normalized and will be in the range
of [0.0, 1.0].
Pupilometry
The diameter of the right eye pupil.
Pupilometry
The value is normalized and will be in the range
of [0.0, 1.0].
69
Pupilometry
The pupilometry module enables measurement of pupil diameters. Pupil diameters are measured at full
frame rate. Pupilometry is an optional module of Smart Eye Pro.
For pupilometry the IR-flashes should be placed in a way such that dark pupils can be guaranteed. This
is done by placing the IR-flashes at least 10 cm from the optical axis of the cameras. Normally the
flashes are placed only a few centimeters on either side of the cameras. Ideally the flashes should be
placed underneath the cameras or 'outside' the cameras if applicable. Avoid placing the flashes on the
upper side of the cameras as this may cause undesired reflections in glasses.
The pupil diameter is measured by matching the image of the dark pupil with all feasible pupil
diameters. This results in a probability function which has its maximum at the best matching diameter.
A probability function is created for each separate eye clip from the cameras. To form the resulting
probability function for one physical eye, the probabilities from all corresponding eye clips are added
together on the maximum of the resulting function.
To determine the consensus pupil diameter for both eyes, the probability functions from the physical
eyes are added together and the maximum of the resulting function defines the diameter.
The quality measure provided with each diameter sample is actually the value of the probability
function at the determined diameter. Quality measures are normalized for each individual and will be a
value in the range [0.0, 1.0].
The set of pupil diameters matched with the image to form the probability function range from 2.5 mm
up to 8 mm with a step of 0.5 mm. This results in a diameter resolution of 0.5 mm. As described below,
a temporal filter is then used to increase the resolution.
Pupil filter
The diameter of the pupil does not change very rapidly. Temporal filtering may therefore be used
successfully to increase the resolution and to reduce the amount of noise.
The pupil filter is a quality weighted average filter. That is, each sample is weighted with its quality and
the average of all weighted samples is calculated. Samples that differ from the average by more than
two standard deviations will be considered to be outliers and will be removed.
Samples within the filter length time are used to produce the filtered value. To reduce delay in output
from the system, only samples calculated before the current time will be used to form the current
filtered value. That is, the filtered value at a given time will only be based on history. One can say that
the filtered value lags from the unfiltered values with about half the filter length.
The default filter length of the pupil filter is 300 ms. This implies that filtered values produced for one
frame corresponds to a frame 150 ms earlier in the output data stream.
The filter length may be adjusted in the Options->Settings->Pupilometry tab.
Quality
Quality figures are always given as a value ranging from 0.0 to 1.0.
Head Position Quality
The head position quality can take 3 values, a value of 1.0 denotes that the head is found and tracked in
at least 2 cameras, while the value 0.5 means that only one camera is tracking. The value 0.0 means that
no face is found at all.
Generally you should only trust a quality value of 1.0.
Gaze Direction Quality
The gaze direction quality reflects the strength of the edge between iris and sclera. This value highly
depends on the eye color, illumination and other factors. Therefore the range of this value is very
individual. In order to be able to use some generic threshold over a number of different individuals an
attempt is made to normalize this value to the range [0.0, 1.0]. The normalized value 0.0 corresponds to
70
the 1st percentile of all collected quality values of the current tracking session while 1.0 corresponds to
the 99th percentile respectively. It takes some samples with good and bad quality measurements before
the normalization has built up enough statistics to be effective. Therefore the quality value of the first
few seconds of a tracking session may not be comparable to the succeeding quality values.
What complicates this further is the fact that normalization is done individually for each camera. This
means that the quality levels from the different cameras may differ slightly, especially in the beginning
of a tracking session.
The quality value that the system puts out is the one of the camera that is expected to have the best view
of the eye. There is no averaging between the cameras. So, when the system thinks another camera has a
better view of the eye, that camera’s quality value is the output. Consequently, if the cameras have
different normalized quality levels, there might be different quality levels put out of the system.
Normally, this effect is only significant at the very beginning of a tracking session. However, if the
system is configured is such way that one of the cameras is used very little, there might be a risk that
this camera will provide quality levels in a different range than the other cameras.
A workaround for this problem is to start each tracking session by looking into each camera once and
blink the eyes. This will fill up the normalization buffer with enough statistics.
71
Diagnosis
We provide a diagnosis value each time there isn't enough information to calculate a gaze vector, or if
the quality of the information is too low to give a correct result.
You get a diagnosis value for each eye. Ideally these values should be 0. All values below 0 indicate
that a gaze vector couldn't be calculated and if they appear often you should examine your system set
up.
Value
Description
Cause
Resolution
0
Gaze Calculated
Successfully.
-1
Less than two detected
irises/pupils.
Less than two irises or
pupils are found in the
cameras for this eye.
Make sure at least two
irises can be seen with
the current camera
position.
-2
Iris/pupil triangulation
failed.
Two or more
irises/pupils are
detected at different
positions.
Check the camera
calibration.
-3
Eyelid opening too
small to accept iris
detection
The iris or pupil is
Check your eyelid
occluded by the eyelids. ranges. Or personal
profile.
-4
No glints found
No reflection of the IR- Check your flashes and
flashes found on the
flash settings. Make
Iris.
sure the lenses are not
out of focus.
-5
Cannot calculate gaze
The gaze direction
calculation failed for
another reason.
-6
Iris/Pupil detection
quality too low
The quality of the
If the iris quality is
detected irises or pupils constantly too low for a
is too low.
subject, try to run the
pupil detection instead.
Contact Smart Eye if
this error persists.
Time Synchronization Specification (Optional module)
The Time Synchronization Module has three main features:
It enables the user to provide a user defined 64 bits value as a timestamp to each frame by writing a
very simple .dll, which is called from the Smart Eye system at capture of each frame. The user
timestamp is passed on to all logs and makes it then very easy to synchronize data from the Smart Eye
system with data from another system utilizing the same timestamps. Instructions on writing the dll
follow below.
The time synchronization module also provides information about latency for each frame, which is the
time from the image capture to the time the log data leaves the application either on file, TCP, UDP or
rs232. This is important in real time applications. Note that this value can and will differ by several 100
microseconds for different data streams, since we wont be sending them at the exact same time.
If the user does not desire to use a user defined time stamp the module will by default provide the realtime timestamp of the frame capture based on the real time clock of the PC. The real-time timestamp is
issued using the Win32 function GetSystemTimeAsFileTime().
72
Writing the dll
For this section it is a good idea to refer to the sample code in the folder TimeService of the Smart Eye
Pro installation, which normally resides in C.\Program files\Smart Eye\Smart Eye Pro x.x\
Compiling the TimeService project results in the default TimeService.dll, which provides the real time
clock of the PC.
The dll has four functions that may be modified by the user. The most important is obviously the
getCurrentTime function, which should return a 64-bit value representing the user time stamp. If
necessary, startup and close down code can be added into the startTimeService and stopTimeService
functions respectively. The timeToString should be able to convert the user time stamp to a readable
value that may be presented in user interface.
The getCurrentTime function of the dll is called upon capture of a video frame from the cameras. The
capture is halted while the waiting for response, thus the getCurrentTime function must return very fast
and must be non-blocking. Ideally only calls to window system dlls should be performed and obviously
no locks or any other pieces of code that may halt the execution can be utilized.
Also, the getCurrentTime function should ideally be reentrant, meaning that more than one thread can
run the function simultaneously without hazards. By default the system expects that the function is
reentrant, but this may be configured in the Options->Settings->Current Timestamp DLL tab.
The sample code also contains a project called “TimeServiceTest”, which is an application for testing
the TimeService.dll outside Smart Eye Pro. The usage of TimServiceTest.exe is:
C:\TimeService\TestDebug\TimeServiceTest.exe <DLLFileName> <offset>
- DLLFileName is the full path to the dll
- offset is an optional time to add to the current time, set this to 0
At run time Smart Eye Pro will set offset to the time elapsed from midpoint of camera exposure to time
of call to getCurrentTime of TimeService.dll.
Logging Specifications
The log file is a tab-separated text file, which contains the data items specified in the format
specification dialog. You can select between a custom format where you pick what you want to log
from a list, or two predefined formats for backward compatibility.
Communication Specifications
TCP/IP socket
The system provides a socket server of type TCP on port 5002 (default). The data is sent in packets. A
packet consists of a header and the actual output of the system. For a more detailed description of the
network packet, please see heading “Packet Specification (TCP/IP, UDP/IP)” below.
Note
Data is sent with network order (high order byte first).
UDP/IP socket
The system can send data packets in UDP format. This option is default off. See heading
“Communication” above for enabling instructions. A packet consists of a header and the actual output
of the system. For an exact definition of the UDP packet, see heading “Packet Specification (TCP/IP,
UDP/IP)” below.
Note
Data is sent with network order (high order byte first).
73
Packet Specification (TCP/IP, UDP/IP)
The basic structure of the data packet produced by the different versions of Smart Eye Pro are the same.
The data packet always starts with the packet header. The packet header declares the type of the packet
and the size of the data included in the packet. Schematically a data packet from Smart Eye Pro looks
like this:
Packet header
Sync Id
SEu32, 4 bytes
Packet type
SEu16, 2 bytes
Packet length
SEu16, 2 bytes
Data
Contents dependent on
packet type
.
.
.
.
FIGURE 38. The basic packet structure of data packets from Smart Eye Pro
Up to version 3.0 the data part in the packet had a fixed length for all packets of the same type. Please
refer to the header file SmartEyePacket.h included in the sample network client included in your Smart
Eye Pro installation for the exact definitions of the data packets up to version 3.0. Packet types for these
packets are 2 or 3.
The output format of Smart Eye Pro 3.0 is quite different from the formats of earlier versions. The
format is not fixed as before, but is composed by a number of tagged sub packets. Please see packet
structure below for an illustration of packet structure.
As before the packet begins with a packet header. The header definition is the same as before, but the
packet type is 4 for the generic packet. After the header a number of sub packets will follow. The size
declared in the packet header indirectly defines how many sub packets there are. That is, the size
defines the size in bytes and not the actual number of sub packets.
The number of sub packets may differ between two consecutive data packets. Client is responsible for
interpreting the data packets correctly.
Each sub packet begins with a sub packet header. The sub packet header contains two pieces of
information: id and size(length). The id indicates what data the sub packet contains, for instance.
SELeftGazeDirection or SEHeadPosition. Please note that the actual data type (double, int e.t.c.) is not
indicated in the packet. The client is expected to look up data type out of the data id. The sample
network client illustrates how this may be done.
Sample network client
The sample client you will find at the same location as the Smart Eye Pro installation. E.g. C:\Program\
SmartEye\Smart Eye Pro 3.0\SampleNetworkClient. The files in this folder are portable and may be
used under Linux. The folder SocketListener contains the project and workspace files for a Visual C++
project. Just open the project file in Visual C++ and build the executable to test the client.
There are two important header files that are needed when creating a network client:
The file SEDataTypes.h contains definitions of the data types that are custom to the Smart Eye System.
The file SEOutputDataIds.h contains definitions of the different data items that the Smart Eye System
may produce. E.g. SEGazeDirection, SETimestamp and SEHeadPosition. The file also contains a
74
lookup table for looking up data type and literal name based on data id. The sample network client
illustrates how this is done.
The vast part of the code resides in the file SocketListener.cpp in the folder SocketListener. If you are
building a client for the generic packet you will only have to consider the cases that handles packet type
4. We strongly recommend all customers to use the generic packet format as this will make the client
more tolerant to future format modifications.
Please note that data is sent in network byte order. That is, on a Intel based system you will need
to reverse the byte order in order to interpret the data correctly. The sample client also
illustrates this.
Packet header
Sync Id
SEu32, 4 bytes
Packet type
SEu16, 2 bytes
Packet length
SEu16, 2 bytes
Subpacket 1
Subpacket
header
Id
SEu16, 2 bytes
Length
SEu16, 2 bytes
Subpacket
data
Data
.
.
.
.
.
Subpacket N
.
.
Subpacket
header
Id
SEu16, 2 bytes
Length
SEu16, 2 bytes
Subpacket
data
Data
.
.
FIGURE 39. The packet structure of new generic format introduced in Smart Eye Pro 3.0
Data types
Data Type
SEu8
SEu16
Consists of
Unsigned integer 1 byte
Unsigned integer 2 bytes
75
Type Alias
SEType_u8
SEType_u16
SEu32
SEs32
SEu64
SEf64
SEPoint2D
SEVect2D
SEPoint3D
SEVect3D
SEString
SEWorldIntersectionStruct
SEWorldIntersection
SEWorldIntersections
Unsigned integer 4 bytes
Signed integer 4 bytes
Unsigned integer 8 bytes
Floating point 8 bytes
SEf64 x
SEf64 y
SEf64 x
SEf64 y
SEf64 x
SEf64 y
SEf64 z
SEf64 x
SEf64 y
SEf64 z
SEu16 size
SEu8 ptr[1024]
SEPoint3D worldPoint
SEPoint3D objectPoint
SEString objectName
SEu16 size = 0..1
SEWorldIntersectionStruct
SEu16 size = 0..N
SEWorldIntersectionStruct
SEType_u32
SEType_s32
SEType_u64
SEType_f64
SEType_Point2D
SEType_Vect2D
SEType_Point3D
SEType_Vect3D
SEType_String
SEType_WorldIntersection
SEType_WorldIntersections
SubPacket Ids
Enum ID
SEFrameNumber
SEEstimatedDelay
SETimeStamp
SEUserTimeStamp
SEFrameRate
SEHeadPosition
SEHeadPositionQ
SEHeadRotationRodrig
ues
SEHeadNoseDirection
SEHeadUpDirection
SEHeadLeftEarDirecti
on
SEHeadHeading
SEHeadPitch
SEHeadRoll
SEHeadRotationQ
SEEyePosition
SEGazeOrigin
SELeftGazeOrigin
SERightGazeOrigin
SEGazeDirection
SEGazeDirectionQ
SELeftEyePosition
SELeftGazeDirection
SELeftGazeDirectionQ
SERightEyePosition
SERightGazeDirection
SERightGazeDirection
Q
SEGazeHeading
SEGazePitch
Enum number
0x0001
0x0002
0x0003
0x0004
0x0005
0x0010
0x0011
0x0012
Data type
SEType_u32
SEType_u32
SEType_u64
SEType_u64
SEType_f64
SEType_Point3D
SEType_f64
SEType_Vect3D
0x0013
0x0014
0x0015
SEType_Vect3D
SEType_Vect3D
SEType_Vect3D
0x0016
0x0017
0x0018
0x0019
0x0020
0x001a
0x001b
0x001c
0x0021
0x0022
0x0023
0x0024
0x0025
0x0026
0x0027
0x0028
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_Point3D
SEType_Point3D
SEType_Point3D
SEType_Point3D
SEType_Vect3D
SEType_f64
SEType_Point3D
SEType_Vect3D
SEType_f64
SEType_Point3D
SEType_Vect3D
SEType_f64
0x0029
0x002a
SEType_f64
SEType_f64
76
SEFilteredGazeDirecti
on
SEFilteredGazeDirecti
onQ
SEFilteredLeftGazeDir
ection
SEFilteredLeftGazeDir
ectionQ
SEFilteredRightGazeDi
rection
SEFilteredRightGazeDi
rectionQ
SEFilteredGazeHeadin
g
SEFilteredGazePitch
SESaccade
SEFixation
SEEyelidOpening
SEEyelidOpeningQ
SELeftEyelidOpening
SELeftEyelidOpeningQ
SERightEyelidOpening
SERightEyelidOpening
Q
SEClosestWorldInterse
ction
SEFilteredClosestWorl
dIntersection
SEAllWorldIntersectio
ns
SEFilteredAllWorldInt
ersections
SEEstimatedGazeOrigi
n
SEEstimatedLeftGaze
Origin
SEEstimatedRightGaze
Origin
SEEstimatedEyePositio
n
SEEstimatedGazeDirec
tion
SEEstimatedGazeDirec
tionQ
SEEstimatedGazeHead
ing
SEEstimatedGazePitch
SEEstimatedLeftEyePo
sition
SEEstimatedLeftGaze
Direction
SEEstimatedLeftGaze
DirectionQ
SEEstimatedLeftGaze
Heading
SEEstimatedLeftGazeP
itch
SEEstimatedRightEyeP
osition
0x0030
SEType_Vect3D
0x0031
SEType_f64
0x0032
SEType_Vect3D
0x0033
SEType_f64
0x0034
SEType_Vect3D
0x0035
SEType_f64
0x0036
SEType_f64
0x0037
0x0038
0x0039
0x0050
0x0051
0x0052
0x0053
0x0054
0x0055
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_f64
SEType_f64
0x0040
SEType_WorldIntersection
0x0041
SEType_WorldIntersection
0x0042
SEType_WorldIntersections
0x0043
SEType_WorldIntersections
0x007a
SEType_Point3D
0x007b
SEType_Point3D
0x007c
SEType_Point3D
0x0080
SEType_Point3D
0x0081
SEType_Vect3D
0x0082
SEType_f64
0x0083
SEType_f64
0x0084
0x0085
SEType_f64
SEType_Point3D
0x0086
SEType_Vect3D
0x0087
SEType_f64
0x0088
SEType_f64
0x0089
SEType_f64
0x008a
SEType_Point3D
77
SEEstimatedRightGaze
Direction
SEEstimatedRightGaze
DirectionQ
SEEstimatedRightGaze
Heading
SEEstimatedRightGaze
Pitch
SEKeyboardState
SELeftDiagnosis
SERightDiagnosis
SELeftGlintsFound
SERightGlintsFound
SECalibrationGazeInte
rsection
0x008b
SEType_Vect3D
0x008c
SEType_f64
0x008d
SEType_f64
0x008e
SEType_f64
0x0056
0x00a0
0x00a1
0x00a2
0x00a3
0x00b0
SEType_String
SEType_f64
SEType_f64
SEType_u32
SEType_u32
SEType_WorldIntersection
Smart Eye Serial Format (RS232)
Up to version 3.0 RS232 was a standard communication interface of Smart Eye Pro. Due to its limited
bandwidth the serial interface is normally not the best choice for communicating data at 60 Hz.
78
CAN Output Specification
The Smart Eye Can Output interface uses 10 predefined packets containing the much of the same
information as in the data output specification.
All packets have a numerical id (nID in the table below) that is added to a user defined message base
for use on the CAN bus. The default message base is 0x400. You will find the message definitions in
SECANMsgTypes.h in the Smart Eye API, where you can also find a sample CANCaseXL-client.
The packet data is normally 8 bytes long, except for the SE_CAN_EYECLOSURE and
SE_CAN_PUPILDIAMETER packets which are 4 bytes each.
To save bandwidth all values (except frame number and user time stamp) are multiplied by 10 000
before they are sent, which means that metric values will be in 1/10 mm, time measurement in 1/10 ms,
and angular values in 1/10 000 radians. Quality values are between 1 and 10000. Most values will be in
the form of 16 bit integers, except for frame number which has to be 32 bit integer and the user defined
time stamp which still is a 64 bit integer.
On the CAN bus, little endian is used, i.e the least significant byte is always sent first for multiple byte
values.
Note
Output Data is always sent as little endian on the CAN bus
Packet ID
nID
Data
Byte Type Id
SE_CAN_SYNCH
1
[0..3] u32 Framenumber
[4..5] u16 Timestamp
[6..7] u16 EstimatedDelay
Unit
Description
1/10ms
1/10ms
You can use the
frame number to
uniquely identify a
frame.
The time stamp will
wrap around at
6.5536 s and the
estimated delay will
saturate at 6.5535s
Used to
synchronize Smart
Eye Pro with
external events.
The head position,
scaled down to 1/10
mm.
SE_CAN_USERTIME
STAMP
2
[0..7] u64 UserTimeStamp
User
Defined
SE_CAN_HEADPOSI
TION
3
SE_CAN_HEADROT
ATION
4
SE_CAN_GAZEORIG
IN
5
[0..1] s16 HeadPosition.x
[2..3] s16 HeadPosition.y
[4..5] s16 HeadPosition.z
[6..7] u16 HeadPositionQ
[0..1] s16 HeadRotation.x
[2..3] s16 HeadRotation.y
[4..5] s16 HeadRotation.z
[6..7] u16 HeadRotation Q
[0..1] s16 GazeOrigin.x
[2..3] s16 GazeOrigin.y
[4..5] s16 GazeOrigin.z
[6..7] u16 GazeOriginQ
1/10 mm
1/10 mm
1/10 mm
0-10000
1/10 mm
1/10 mm
1/10 mm
0-10000
1/10 mm
1/10 mm
1/10 mm
0-10000
79
The head position
in rodrigues format,
scaled down to
1/10000 radians.
The origin of the
gaze vector, scaled
down to 1/10 mm.
SE_CAN_GAZEDIRE
CTION
6
[0..1] s16 GazeDirection.x
[2..3] s16 GazeDirection.y
[4..5] s16 GazeDirection.z
[6..7] u16 GazeDirectionQ
[0..1] s16 EyeLidOpening
[2..3] u16 EyeLidOpeningQ
+-10000
+-10000
+-10000
0-10000
1/10 mm
0-10000
SE_CAN_EYECLOS
URE
7
SE_CAN_BOTHEYE
CLOSURES
8
[0..1] s16 LeftEyeLidOpening
[2..3] u16 LeftEyeLidOpeningQ
[4..5] s16 RightLidOpening
[6..7] u16 RightLidOpeningQ
1/10 mm
0-10000
1/10 mm
0-10000
SE_CAN_PUPILDIA
METER
9
[0..1] s16 Pupildiameter
[2..3] u16 PupildiameterQ
1/10 mm
0-10000
SE_CAN_BOTHPUPI
LDIAMETERS
10
[0..1] s16 LeftPupildiameter
[2..3] u16 LeftPupildiameterQ
[4..5] s16 RightPupildiameter
[6..7] u16 RightPupildiameterQ
1/10 mm
0-10000
1/10 mm
0-10000
80
The unit vector
gaze direction, but
between -10000
and 10000
The average eye lid
opening value. Unit
is 1/10 mm.
Both the left and
right eye lid
opening values with
quality values. Unit
is 1/10 mm.
The diameter of the
pupil in 1/10 mm
scale.
The diameter of the
pupil in 1/10 mm
scale for both the
left and the right
eye.
Contact Information
Smart Eye AB
Första Långgatan 28 B
SE - 413 27 Göteborg
Sweden
Tel
+46 31 60 61 60
Fax
+46 31 701 05 15
[email protected]
www.smarteye.se
81